Daily reading queue

150 items for 2026-05-29 across 3 categories.

Active sources: 7. Sources represented in this queue: 6. The cron runs daily at 06:00 server time; arxiv RSS is often empty on weekends.

Linked to your results

score 100arxiv stat.ML (Machine Learning)arxiv:2605.29645unread
The Sample Complexity of Multiclass and Sparse Contextual Bandits
Liad Erez, Fan Chen, Alon Cohen, Tomer Koren, Yishay Mansour, Shay Moran, Alexander Rakhlin · 2026-05-29
arXiv:2605. 29645v1 Announce Type: cross Abstract: We study contextual bandits in the stochastic i.
Read next because The Sample Complexity of Multiclass and Sparse Contextual Bandits overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, class, rect, rate, compare, factor, contexts, lora. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29645v1 Announce Type: cross Abstract: We study contextual bandits in the stochastic i.i.d.\ setting, where a learner observes contexts drawn from an unknown distribution, selects actions from a finite set $A$, and aims to identify an approximately optimal policy from a given class based on bandit feedback. Motivated by bandit multiclass classification with zero-one rewards, we focus on the \emph{$s$-sparse} setting in which, for every context, the reward vector has $L_1$-norm at most $s \ll |A|$. Our main result is the design of algorithms that, with high probability, output an $\epsilon$-optimal policy compared to policy class $\Pi$ using $\tilde{O} ((s/\epsilon^2 + |A|/\epsilon)\log |\Pi|/\delta)$ samples. We extend this bound to general Natarajan classes and complement it with a matching lower bound (up to logarithmic factors), thereby closing a substantial gap left by prior work (Erez et al., 2024, 2025), which incurred an additional $\Theta(|A|^9)$ dependence. We obtain these results via two complementary approaches. First, we analyze contextual bandits through the lens of contextual decision making with structured observations, designing an exploration-by-optimization algorithm whose sample complexity is governed by the \emph{decision-estimation coefficient} (DEC; Foster et al., 2021, 2022). We show that, with $s$-sparse rewards, the induced model class admits a sharp DEC bound that scales with $s$ and directly yields the optimal rate. Since this approach is largely information-theoretic and involves solving complex min-max optimization problems, we also develop a second, more specialized algorithmic method based on a low-variance exploration technique. This approach leads to concrete, tractable algorithms and naturally extends to contextual combinatorial semi-bandits, leading to improved sample complexity guarantees for bandit multiclass list classification.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29387unread
On the Optimizer Dependence of Neural Scaling Laws
Vansh Ramani, Shourya Vir Jain · 2026-05-29
arXiv:2605. 29387v1 Announce Type: cross Abstract: The scaling exponent $\alpha$ in neural scaling laws $L(N) \propto N^{-\alpha}$ is commonly treated as a fixed constant set by architecture and data.
Read next because On the Optimizer Dependence of Neural Scaling Laws overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone", experiment "Follow-up to #354: cascading chunk-binding — does A→B, B→C, C→D propagate the full chain on a recipient trained only to emit A?". Matching terms: alpha, control, full, test, language, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29387v1 Announce Type: cross Abstract: The scaling exponent $\alpha$ in neural scaling laws $L(N) \propto N^{-\alpha}$ is commonly treated as a fixed constant set by architecture and data. We present evidence that $\alpha$ depends systematically on the optimizer. In controlled random-feature regression experiments -- the canonical theoretical framework for neural scaling -- we measure $\alpha$ across five optimizer variants and six spectral conditions. Preconditioned optimizers consistently yield steeper scaling (larger $\alpha$), with the $\alpha$-shift increasing across most of the tested spectral range, peaking near $s = 1.5$, and remaining large at $s = 2.0$. At $s \approx 1.0$ (characteristic of natural language), the full natural gradient achieves $\alpha \approx 0.31$ versus $\alpha \approx 0.12$ for gradient descent -- a $2.6\times$ larger fitted exponent that, within the random-feature model, compounds with each model-size doubling. Whether and how this exponent shift transfers to large-scale LLM training -- where recent evidence suggests the advantage may attenuate with scale -- remains an important open question. Our results imply that scaling-law forecasts should account for optimizer choice, and we provide a spectral diagnostic predicting when advanced optimizers will pay off.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29351unread
Attention as In-Context Empirical Bayes: A Two-Stage View via Particle Dynamics
Matthew Smart, Soumya Ganguly, Nilava Metya, Alexandre V. Morozov, Anirvan M. Sengupta · 2026-05-29
arXiv:2605. 29351v1 Announce Type: cross Abstract: We study minimal attention-only transformers under all-token corruption and show they admit a two-stage empirical Bayes interpretation.
Read next because Attention as In-Context Empirical Bayes: A Two-Stage View via Particle Dynamics overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, class, under, width, token, without, stage, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29351v1 Announce Type: cross Abstract: We study minimal attention-only transformers under all-token corruption and show they admit a two-stage empirical Bayes interpretation. A single attention step computes a kernel-weighted posterior mean with respect to the empirical distribution defined by the context. Depth refines this distribution through particle dynamics (Stage 1), while a long-range skip-connection carries the noisy input as a query for posterior inference (Stage 2), revealing distinct statistical roles for depth and attention residuals. The framework isolates a minimal setting in which the context itself induces a depth-dependent energy landscape governing in-context inference. We show that effective denoising can emerge without an explicit noise schedule: a fixed kernel bandwidth and finite integration horizon suffice, yielding a principled depth-noise relationship. We further establish a posterior-mean recovery guarantee for a class of well-behaved priors, where the empirical estimator converges to the Bayes-optimal predictor under asymptotic conditions. Connecting these dynamics to reverse-diffusion limits, our results provide a statistical interpretation of attention as in-context inference via sample-based posterior estimation, without explicit density modeling.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29189unread
Bayesian Multiplicity Correction in the Probabilistic Forward Stepwise Framework
Andrew Womack, Daniel Taylor-Rodriguez · 2026-05-29
arXiv:2605. 29189v1 Announce Type: cross Abstract: We develop a natural Bayesian multiplicity-correcting prior distribution within the probabilistic forward stepwise representation of model space priors for regression problems.
Read next because Bayesian Multiplicity Correction in the Probabilistic Forward Stepwise Framework overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: rect, correct, compare, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29189v1 Announce Type: cross Abstract: We develop a natural Bayesian multiplicity-correcting prior distribution within the probabilistic forward stepwise representation of model space priors for regression problems. The proposed prior, obtained from making an analogy to the Holm procedure, exhibits behavior closely aligned with that of the Matryoshka doll prior. We compare both priors to several other priors, including some recently put forward as objective choices for model space prior probabilities. Our comparisons indicate that adequate multiplicity correction requires a degree of sparsity that many recommended priors do not provide, and we argue that multiplicity correction itself offers a principled and transparent criterion for specifying model space priors in regression.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29148unread
Optimal Gap-Dependent Regret for Private Stochastic Decision-Theoretic Online Learning
Tommaso Cesari, Roberto Colomboni · 2026-05-29
arXiv:2605. 29148v1 Announce Type: cross Abstract: We study stochastic decision-theoretic online learning with full information and event-level pure differential privacy.
Read next because Optimal Gap-Dependent Regret for Private Stochastic Decision-Theoretic Online Learning overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, soft, prefix, line, rate, control, full, length. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29148v1 Announce Type: cross Abstract: We study stochastic decision-theoretic online learning with full information and event-level pure differential privacy. A COLT open problem of Hu and Mehta asks to determine the optimal gap-dependent regret rate for stochastic decision-theoretic online learning under pure event-level differential privacy. For $K$ actions, losses in $[0,1]$, and a unique best action separated from the second-best action by gap $\Delta_{\min}$, the known lower bound is of order $ \frac{\log K}{\min\{\Delta_{\min},\varepsilon\}}, $ or equivalently, up to universal constants, of order \[ \frac{\log K}{\Delta_{\min}}+\frac{\log K}{\varepsilon}. \] We give a horizon-free pure-DP algorithm and prove the explicit regret bound \[ \operatorname{Reg}_T \le 1000 \cdot \left(\frac{\log K}{\Delta_{\min}}+\frac{\log K}{\varepsilon}\right) \] for every horizon $T$. The numerical constant is not optimized. The algorithm partitions time into blocks of exponentially increasing size, plays a single action throughout each block, and chooses the next action by an exponential mechanism applied to a data-independent random prefix of the previous block. The random prefix converts block regret into a sum, over all prefix lengths, of softmax selection errors. A single entropy-potential argument controls all privacy-dominated large-gap actions at cost $\log K/\varepsilon$.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.30319unread
Improved Guarantees for Heterogeneous Treatment-Effect Estimation via Matrix Completion
Anay Mehrotra, Phuc Tran, Van H. Vu, Manolis Zampetakis · 2026-05-29
arXiv:2605. 30319v1 Announce Type: new Abstract: A central goal of modern causal inference is estimating heterogeneous treatment effects to answer questions like "how does an intervention affect each unit," rather than only on average.
Read next because Improved Guarantees for Heterogeneous Treatment-Effect Estimation via Matrix Completion overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, good, line, rate, without, does, completion. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.30319v1 Announce Type: new Abstract: A central goal of modern causal inference is estimating heterogeneous treatment effects to answer questions like "how does an intervention affect each unit," rather than only on average. We study this problem with panel-data where we observe $n$ units across $m$ times under unknown, non-uniform treatment assignments. The data in this setting is naturally represented as a matrix of all unit--time treatment effects. Estimating heterogeneous treatment effects can then be expressed as obtaining a good estimation of each row's average in this matrix. This allows us to formulate the problem as matrix completion, which can be solved under natural low-rankness assumptions. However, existing matrix-completion guarantees are not powerful enough to get meaningful bounds for the per-row guarantee required for estimating the heterogeneous treatment effect; roughly speaking, they are only useful for estimating average treatment effect bounds, as also illustrated in a recent line of work. We give a simple, computationally efficient estimator that, without knowledge of the propensities and under standard low-rankness and regularity assumptions, achieves a row-wise $\ell_2$ error of $\tilde{O}(\sqrt{\frac{1}{n} + \frac{n}{m^2}})$. Technically, our analysis establishes the first sharp row-wise $\ell_2$-perturbation bound for low-rank approximation, complementing existing spectral-, Frobenius-, and entrywise perturbation theory.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.30292unread
Leave a Window Out: Modifying the Jackknife for Predictive Inference in Time Series
Hanyang Jiang, Rina Foygel Barber, Ashwin Pananjady, Yao Xie · 2026-05-29
arXiv:2605. 30292v1 Announce Type: new Abstract: Conformal prediction methods enjoy strong theoretical and empirical predictive inference performance, provided the data is exchangeable, and predictors are trained in a memoryless fashion.
Read next because Leave a Window Out: Modifying the Jackknife for Predictive Inference in Time Series overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rate, trained, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.30292v1 Announce Type: new Abstract: Conformal prediction methods enjoy strong theoretical and empirical predictive inference performance, provided the data is exchangeable, and predictors are trained in a memoryless fashion. However, these assumptions and constraints are impractical in many real-data settings, such as time series (where temporal dependence violates exchangeability, and where memoryless predictors will inevitably have poor predictive accuracy). Recent work shows that the split conformal prediction method is robust to these issues of memory-based predictors and deviations from exchangeability that are common features of time-series data. However, since using sample splitting can lead to lower accuracy, this motivates asking whether other predictive inference methods (that do not rely on data splitting) could also be reliably used in the time series setting. In this work, we show that the vanilla leave-one-out jackknife can suffer an arbitrary loss of coverage even in canonical time series models with mild temporal dependence. As a remedy, we propose a careful modification tailored to such settings, which we term the \emph{leave-a-window-out} (LWO) method, and show that it can achieve valid coverage provided that the model-fitting procedure satisfies mild stability properties. Our proofs are based on quantifying the degree to which the data departs from \emph{cyclic exchangeability}, and we introduce new coefficients to measure the extent of this departure. Experiments on time series data demonstrate that our LWO method often enjoys valid coverage when the vanilla jackknife fails to cover, while producing much narrower intervals than split conformal prediction.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.30167unread
Visual Spatial Learning: Single-Field Spatial Interpolation Using Convolutional Neural Networks
Daniel Tinoco, Raquel Menezes, Carlos Baquero, Alexandra Silva · 2026-05-29
arXiv:2605. 30167v1 Announce Type: new Abstract: Predicting a complete spatially correlated field from sparse observations is a fundamental challenge in spatial statistics and environmental modelling.
Read next because Visual Spatial Learning: Single-Field Spatial Interpolation Using Convolutional Neural Networks overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: class, rect, under, rate, without, does, trained, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.30167v1 Announce Type: new Abstract: Predicting a complete spatially correlated field from sparse observations is a fundamental challenge in spatial statistics and environmental modelling. Classical interpolation methods such as Kriging rely on Gaussian process assumptions and variography, which can limit their effectiveness in non-stationary settings and require substantial domain expertise. In this work, we leverage an architecture based on convolutional neural networks (CNNs) for spatial interpolation that is trained and applied on a single partially observed field, without access to external data or prior fields. The model is supervised directly on the observed locations and learns to predict values at unobserved points on the user defined grid. Unlike Kriging, our method does not require explicit covariance modelling or variogram estimation, and it can flexibly capture local spatial patterns in a data-driven manner. This work demonstrates the potential of CNNs for single-instance spatial interpolation under sparse supervision, offering a practical alternative to classical geostatistical methods, and extending the use of CNNs to a new problem domain.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.30153unread
Diffusion Models Are Statistically Optimal for Learning Low-Dimensional Multi-Modal Distributions
Jingda Wu, Changxiao Cai · 2026-05-29
arXiv:2605. 30153v1 Announce Type: new Abstract: Score-based diffusion models have demonstrated remarkable empirical success in learning high-dimensional distributions, particularly those exhibiting low-dimensional and multi-modal structures.
Read next because Diffusion Models Are Statistically Optimal for Learning Low-Dimensional Multi-Modal Distributions overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, under, rate, without, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.30153v1 Announce Type: new Abstract: Score-based diffusion models have demonstrated remarkable empirical success in learning high-dimensional distributions, particularly those exhibiting low-dimensional and multi-modal structures. However, theoretical understanding of their statistical efficiency remains limited. Existing theories typically rely on strong regularity assumptions, such as uniformly bounded densities or globally smooth score functions, which fail to capture such intrinsic structures. In this work, we study the sample complexity of diffusion models for learning distributions supported on a union of low-dimensional subspaces. Assuming that the data distribution within each subspace is subgaussian, we show that diffusion models require at most $\widetilde{O}(\varepsilon^{-k \vee 2})$ samples to achieve $\varepsilon$ error in 1-Wasserstein distance, where $k$ is the intrinsic dimension. This near-optimal convergence rate depends only on the intrinsic dimension and significantly improves upon prior theoretical guarantees that suffer from the curse of dimensionality. Notably, our analysis applies to a broad collection of distributions without imposing smoothness, bounded-density, or log-concavity assumptions. Overall, our results show that diffusion models can statistically adapt to intrinsic low-dimensional structure while naturally accommodating multi-modal data, offering a rigorous theoretical justification for their success in complex high-dimensional learning tasks.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29669unread
Eigen-Spike Emergence and Quadratic Equivalents for Conjugate Kernels on Nonlinearly Separable Data
Collin Cranston, Zhichao Wang, Todd Kemp, Michael W. Mahoney · 2026-05-29
arXiv:2605. 29669v1 Announce Type: new Abstract: Recent work in random matrix theory (RMT) has developed the notion of deterministic equivalents: typically linear surrogate models that approximate the spectral behavior of large nonlinear random matrices, such as nonlinear feature maps in neural networks (NNs).
Read next because Eigen-Spike Emergence and Quadratic Equivalents for Conjugate Kernels on Nonlinearly Separable Data overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)". Matching terms: class, under, line, trained, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29669v1 Announce Type: new Abstract: Recent work in random matrix theory (RMT) has developed the notion of deterministic equivalents: typically linear surrogate models that approximate the spectral behavior of large nonlinear random matrices, such as nonlinear feature maps in neural networks (NNs). On the one hand, these deterministic equivalents make theoretical predictions tractable by reducing a complex model to a simpler model with properties that fall under the umbrella of classical RMT tools. However, this leaves open the question of whether this idealized linear equivalence remains meaningful when dealing with high-dimensional nonlinearly separable data, such as performing clssification on nonlinearly separable data. Motivated by this, we consider the conjugate kernel (CK), which is the nonlinear feature map of a feedforward NN, under a canonical nonlinearly separable dataset, the XOR problem; and we use the study of informative outlier eigenvalues in the CK and whether their corresponding eigenvectors asymptotically align with XOR labels as a proxy for nonlinear learnability. We develop a robust quadratic equivalent to the spiked CK matrix that enables a precise analysis of emergent informative spikes, as one modifies various knobs common in ML practice: sample complexity, signal-to-noise ratio (SNR), nonlinear activation choice, and pretrained features. In each of these scenarios, we derive a precise BBP-type phase transition in which linear classification via the CK eigenvectors becomes possible. Our analysis helps translate the power of deterministic equivalence tools in RMT to study problems of practical relevance in ML.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29642unread
Matching Rates and Optimal Allocation for Federated Probe-Logit Distillation under Heterogeneous Bandwidth Budgets
Prasanjit Dubey, Xiaoming Huo · 2026-05-29
arXiv:2605. 29642v1 Announce Type: new Abstract: In federated language modeling, $K$ nodes each hold $n$ samples but cannot pool data or exchange full-precision gradients or weights.
Read next because Matching Rates and Optimal Allocation for Federated Probe-Logit Distillation under Heterogeneous Bandwidth Budgets overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)". Matching terms: fill, under, width, token, line, rate, full, language. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29642v1 Announce Type: new Abstract: In federated language modeling, $K$ nodes each hold $n$ samples but cannot pool data or exchange full-precision gradients or weights. We study the minimax rate at which a conditional distribution over $V$ tokens can be estimated when each node may upload at most $B$ bits per query in a public probe set. In federated probe-logit distillation (FPLD), each node transmits a scalar-quantized logit vector on the probe set, and an aggregator distills a global parametric student. Prior work (Dubey and Huo, 2026) establishes a high-probability KL rate $O(d/(Kn) + \rho\sqrt{V \log V / m} + K^{-1} \cdot 2^{-2B/V})$ plus optimization slack, with the bandwidth term in its trace-sharpened form. Whether this bandwidth-term rate is tight, and how the upper bound generalizes to heterogeneous per-node bandwidths, are left open. We close both gaps. First, the dithered FPLD construction has a matching single-round lower bound $\Omega(K^{-1} \cdot 2^{-2B/V})$ under non-degeneracy, pinning the bandwidth-axis rate at $\Theta(K^{-1} \cdot 2^{-2B/V})$. $T$-round sequential refinement with nested/scaled residual quantizers achieves $O(K^{-1} \cdot 2^{-2TB/V})$; vanilla FPLD's $T$-independent bandwidth term is suboptimal for every $T > 1$. Second, we establish a heterogeneous-bandwidth upper bound for per-node budgets $B_i$, paired with a closed-form optimal allocation $B_i^* = B_{\mathrm{tot}}/K + (V/2) \log_2(w_i / \bar{w}_g)$, a log-tilted water-filling rule that is the per-node analogue of reverse water-filling for distortion-rate optimization. A plug-in adaptive variant estimates the weights from a short warm-up phase and attains $1 + O(\sqrt{\log(K/\delta)/(m T_0)})$ relative suboptimality. Synthetic n-gram simulations confirm that empirical KL is bracketed by the upper and lower bounds and that the optimal allocation strictly dominates uniform and inverse-weighted baselines under heterogeneous clipping.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.28961unread
Dynamics of Stochastic Momentum with Sparse Updates in High Dimensions
Katie Everett, Elliot Paquette · 2026-05-29
arXiv:2605. 28961v1 Announce Type: new Abstract: Existing theory of momentum assumes that gradients arrive at every parameter at a roughly constant rate, an assumption violated in practice by heavy-tailed data distributions and modern architectures.
Read next because Dynamics of Stochastic Momentum with Sparse Updates in High Dimensions overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)". Matching terms: class, under, token, rate, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.28961v1 Announce Type: new Abstract: Existing theory of momentum assumes that gradients arrive at every parameter at a roughly constant rate, an assumption violated in practice by heavy-tailed data distributions and modern architectures. We theoretically analyze the dynamics of two tractable models of momentum under sparse updates: a least squares model with sparse inputs and a logistic regression model with a rare class. Both admit exact closed-form second-moment dynamics whose high-dimensional limits we characterize across three scaling exponents for sparsity, batch size, and momentum decay. The phase structure on both problems is governed by the ratio of two intrinsic timescales: a momentum retention timescale (how many active updates the buffer survives) and a learning timescale (how many active updates it takes to reduce the squared error). When learning is much slower than retention, the limit matches SGD; when learning is faster, the system is unstable; where the timescales coincide, we recover classical heavy-ball dynamics. The oscillatory dynamics occur at different momentum values for different token sparsity, creating a spectral conflict for global momentum across token frequencies.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29524unread
KBF: Knowledge Boundary as Fingerprint for Language Model and Black-Box API Auditing
Yijia Fang, Yiqing Feng, Bingyu Li, Mingxun Zhou · 2026-05-29
arXiv:2605. 29524v1 Announce Type: new Abstract: Relay and reseller APIs increasingly intermediate access to large language models (LLMs), but users have no direct way to verify that a claimed endpoint is actually serving the advertised model.
Read next because KBF: Knowledge Boundary as Fingerprint for Language Model and Black-Box API Auditing overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)", clean result "A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)". Matching terms: rect, under, rate, control, without, language, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29524v1 Announce Type: new Abstract: Relay and reseller APIs increasingly intermediate access to large language models (LLMs), but users have no direct way to verify that a claimed endpoint is actually serving the advertised model. We introduce KBF, a low-cost black-box auditing protocol that fingerprints model APIs using stable numerical recall near the knowledge boundary. Across 16 production LLM endpoints, KBF flags all 155 economically relevant substitutions without rejecting any same-model controls, remains stable under deployment variation, detects high-separation mixed-routing attacks when only 5-10% of traffic is substituted, and finds that 7 of 27 platform model cells in a six-platform shadow API audit are statistically inconsistent with their reference endpoints, with inconsistencies concentrated on premium Claude endpoints.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29353unread
DeepFake Forensics AI: A Multi-Modal Detection and Blockchain-Anchored Evidence Management Platform
Naisha Minnah · 2026-05-29
arXiv:2605. 29353v1 Announce Type: new Abstract: The proliferation of AI-generated synthetic media poses a critical threat to the integrity of digital evidence in legal and forensic contexts.
Read next because DeepFake Forensics AI: A Multi-Modal Detection and Blockchain-Anchored Evidence Management Platform overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, rect, rate, control, chain, contexts. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29353v1 Announce Type: new Abstract: The proliferation of AI-generated synthetic media poses a critical threat to the integrity of digital evidence in legal and forensic contexts. Existing deepfake detection systems typically address a single modality and provide no mechanism for tamper-proof evidence preservation. We present DeepFake Forensics AI, a unified platform that detects synthetic media across image, video, and audio modalities, identifies generative architecture fingerprints, and anchors forensic evidence immutably on the Ethereum blockchain. Our system trains four independent neural networks from scratch: an EfficientNet-B4 image detector (AUC = 0.9868), a Bidirectional LSTM video detector (AUC= 0.9628), an ECAPA-TDNN audio detector (EER = 18.63%), and a novel GAN fingerprinting module (accuracy = 99.88%) that identifies the generative architecture behind a fake image. Evidence files are hashed with SHA-256, stored on IPFS via Pinata, and registered on-chain via a Solidity smart contract with role-based access control. The platform provides a React frontend and FastAPI backend suitable for deployment in forensic and legal workflows. To our knowledge, this is the first system to unify multi-modal deepfake detection with blockchain-based chain-of custody management.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29226unread
S3C2 Summit 2025-09: Industry Secure Supply Chain Summit
Md Atiqur Rahman, Yasemin Acar, Michel Cucker, William Enck, Alexandros Kapravelos, Christian Kastner, Dominik Wermke, Laurie Williams · 2026-05-29
arXiv:2605. 29226v1 Announce Type: new Abstract: Today's digital ecosystem relies heavily on software supply chains, which enable developers to reuse code and ship software at scale.
Read next because S3C2 Summit 2025-09: Industry Secure Supply Chain Summit overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, rect, soft, source, rate, full, chain. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29226v1 Announce Type: new Abstract: Today's digital ecosystem relies heavily on software supply chains, which enable developers to reuse code and ship software at scale. However, a single vulnerable component can jeopardize the entire supply chain. In recent years, cyberattacks in software supply chains have become increasingly common. These attacks can disrupt critical systems and put organizations, including major software companies, government agencies, and open-source contributors, at risk. This growing threat has led to increased attention from both the software industry and the U.S. government toward strengthening software supply chain security. On September 15, 2025, three researchers from the NSF-backed Secure Software Supply Chain Center (S3C2) convened a Secure Software Supply Chain Summit, bringing together 10 practitioners from 8 organizations across diverse domains. The goals of the Summit were threefold: (1) to facilitate cross-industry sharing of practical experiences and challenges in securing software supply chains; (2) to foster new collaborations among participants; and (3) to identify pressing challenges to guide future research directions. The Summit featured discussions on six central topics: vulnerable dependencies, component and container choice, malicious commits, build infrastructure, culture, and the role of LLMs in the supply chain. For each topic, participants engaged with a curated set of discussion questions designed to gather insights and pain points. This report summarizes the key takeaways from these discussions. Each section highlights which topics continued from previous summits and which ideas emerged for the first time in this summit; the full list of initial discussion prompts is provided in the appendix.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29210unread
SAMD: A Tool for Identifying False Data Injection Scenarios in AI/ML-enabled Medical Devices
Mohammadreza Hallajiyan, Xueren Ge, Athish Pranav Dharmalingam, Gargi Mitra, Shahrear Iqbal, Homa Alemzadeh, Karthik Pattabiraman · 2026-05-29
arXiv:2605. 29210v1 Announce Type: new Abstract: The growing integration of artificial intelligence (AI) and machine learning (ML) in medical systems requires effective measures to address emerging security risks.
Read next because SAMD: A Tool for Identifying False Data Injection Scenarios in AI/ML-enabled Medical Devices overlaps with clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone". Matching terms: wrong, rate, control, language, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29210v1 Announce Type: new Abstract: The growing integration of artificial intelligence (AI) and machine learning (ML) in medical systems requires effective measures to address emerging security risks. One such risk is that of adversaries introducing false data through vulnerable system components during inference, causing misdiagnosis and wrong treatments. These risks are challenging to anticipate and address in the design phase, as the system assembly partially occurs during actual use by end users. To address this concern, we introduce SAMD, an automated tool for performing System Theoretic Process Analysis for Security (STPA-Sec) on AI/ML-enabled medical devices during the design phase. SAMD models the medical system as a control structure, treating all system components as potential points for injecting false data into the ML engine. It leverages state-of-the-art vulnerability databases and Large Language Models (LLMs) to automate vulnerability discovery and generate a list of potential attack scenarios. We demonstrate SAMD's effectiveness through case studies on five FDA-cleared medical devices, showcasing its ability to identify vulnerable points and potential attack paths. We find that SAMD has 100% precision in identifying target device technologies in the case studies' documents, retrieves the known vulnerabilities linked to them (with 63.2% precision), and generates highly relevant attack scenarios on the ML model, including detailed steps that an adversary might take (with 95.3% accuracy, and the highest time taken being 191.64s).
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29140unread
S3C2 Summit 2025-07: Government Secure Supply Chain Summit
Sivana Hamer, Pat Morrison, William Enck, Yasemin Acar, Michel Cukier, Alexandros Kapravelos, Christian K\"astner, Dominik Wermke, Laurie Williams · 2026-05-29
arXiv:2605. 29140v1 Announce Type: new Abstract: Software supply chains, while providing immense economic and software development value, are only as strong as their weakest link.
Read next because S3C2 Summit 2025-07: Government Secure Supply Chain Summit overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rect, soft, source, chain, language, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29140v1 Announce Type: new Abstract: Software supply chains, while providing immense economic and software development value, are only as strong as their weakest link. Over the past several years, there has been an exponential increase in cyberattacks specifically targeting vulnerable links in critical software supply chains. The attacks disrupt day-to-day functioning and threaten the security of nearly everyone on the internet, from billion-dollar companies and government agencies to hobbyist open-source developers. The evolving threat of software supply chain attacks has garnered interest from both the software industry and governments worldwide in improving software supply chain security. On Thursday, July 9th, 2025, 3 researchers from the NSF-backed Secure Software Supply Chain Center (S3C2) conducted a Secure Software Supply Chain Summit with a diverse set of 12 participants from 6 US government agencies. The goals of the Summit were: (1) to enable sharing between participants from different industries regarding practical experiences and challenges with software supply chain security; (2) to help form new collaborations; and (3) to learn about the challenges facing participants to inform our future research directions. The summit consisted of discussions of six topics relevant to the government agencies represented, including software bill of materials (SBOMs); compliance; malicious commits; build infrastructure; culture; and large language models (LLMs) and security. For each topic of discussion, we presented participants with a list of questions to spark conversation and an overview of the discussions of two industry summit held in the past year. In this report, we provide a summary of the summit. The initial discussion questions for each topic are provided in the appendi
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29131unread
Techreport: Evaluating Tor-based Location Privacy for Ethereum Validators
Muhammad Umar Janjua, Akshaya Mani, U\u{g}ur \c{S}en, Daniel Kaiser · 2026-05-29
arXiv:2605. 29131v1 Announce Type: new Abstract: Privacy and anonymity of validators, especially regarding IP address linkability, are essential to protect the Ethereum network from various attacks.
Read next because Techreport: Evaluating Tor-based Location Privacy for Ethereum Validators overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: latin, eval, rate, implement, without, chain, test. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29131v1 Announce Type: new Abstract: Privacy and anonymity of validators, especially regarding IP address linkability, are essential to protect the Ethereum network from various attacks. Network-level attacks, such as DoS, can interrupt validators and affect the overall security of the Ethereum network. Correlating the IP addresses of validators with their identities, along with knowledge about their action slots can be exploited by attackers to cause network delays, MEV exploitation, and finality risks. Therefore, ensuring the unlinkability of a validator's IP and identity is crucial for maintaining the network's trust and resilience. In this techreport, we first provide a review of the existing network and consensus layer techniques that have been proposed for maintaining validator privacy in the Ethereum blockchain. Secondly, we evaluate a Tor-based protocol named Tor push that helps unlink validator identities (IDs) from their nodes' IP addresses, thereby making it difficult to determine any end-to-end correlation between validator IDs and IP addresses of validators' beacon nodes. To evaluate the effectiveness of Tor push, we present a working, deployed proof-of-concept (PoC) implementation in the Nimbus Ethereum client. Our PoC deployment pushes attestations, aggregations, and block proposals over Tor to the Goerli testnet. Furthermore, we also analyse the security and latency of Tor push. Our experimental results suggest that Tor can be incorporated into the existing Ethereum network with a tolerable latency overhead of 613.82 ms on average and without compromising the overall network performance while enhancing the location privacy of validators in the Ethereum network.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.28999unread
Measuring Real-World Prompt Injection Attacks in LLM-based Resume Screening
Mohan Zhang, Yuqi Jia, Zhen Tan, Steven Jiang, Neil Zhenqiang Gong, Tianlong Chen, Dawn Song · 2026-05-29
arXiv:2605. 28999v1 Announce Type: new Abstract: LLMs are vulnerable to prompt injection attacks.
Read next because Measuring Real-World Prompt Injection Attacks in LLM-based Resume Screening overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, eval, rate, full, screen. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.28999v1 Announce Type: new Abstract: LLMs are vulnerable to prompt injection attacks. However, this vulnerability has been primarily demonstrated conceptually in academic studies or through a few anecdotal case studies. Its prevalence and impact in real-world LLM-based applications are largely unexplored. In this work, we present the first systematic study of prompt-injection attacks in a widely used application: LLM-based resume screening. Our analysis is based on approximately 200K real-world resumes collected over multiple years by hireEZ. We first design tailored methods to detect prompt injection in resumes. Manual validation on a small-scale dataset demonstrates that our detectors achieve high precision and outperform state-of-the-art general-purpose detectors. We then apply our detector to the full resume dataset and conduct a comprehensive measurement study of real-world prompt injection attacks. Our analysis reveals several intriguing findings: approximately 1% of resumes contain hidden prompt injections; the prevalence of such injected resumes has increased noticeably over the past one to two years; and more than 90% of injected prompts do not use explicit instructions. These results provide the first evidence of large-scale prompt injection in real-world LLM-based applications and lay the groundwork for future studies to understand and mitigate such attacks.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29229unread
Tailoring the Curriculum: Student-Centered Reasoning Distillation via Dynamic Data-Model Compatibility
Jiahao Huang, Fei Cheng, Junfeng Jiang, Akiko Aizawa · 2026-05-29
arXiv:2605. 29229v1 Announce Type: new Abstract: Reasoning distillation transfers complex reasoning abilities from large language models (LLMs) to smaller ones, yet its success depends on how well the training data align with the student model.
Read next because Tailoring the Curriculum: Student-Centered Reasoning Distillation via Dynamic Data-Model Compatibility overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rate, capability, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29229v1 Announce Type: new Abstract: Reasoning distillation transfers complex reasoning abilities from large language models (LLMs) to smaller ones, yet its success depends on how well the training data align with the student model. This paper introduces the Data-Model Compatibility (DMC) metric, which can be used to assess the suitability of a dataset for reasoning distillation on a student model. DMC provides an assessment by jointly considering data quality, relative difficulty, and student capability. We validated the effectiveness of DMC from two perspectives: (1) DMC exhibits a strong correlation with reasoning distillation performance; and (2) using DMC as the criterion for data selection leads to improved reasoning distillation performance. Both findings are consistently demonstrated across multiple student models and tasks. Moreover, since the DMC of each dataset dynamically changes during training, our experiments demonstrate that dynamically selecting datasets based on DMC can further enhance performance.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29168unread
Better Later Than Sooner: Neuro-Symbolic Knowledge Graph Construction via Ontology-grounded Post-extraction Correction
Lorenzo Loconte, Timothy Hospedales, Cristina Cornelio · 2026-05-29
arXiv:2605. 29168v1 Announce Type: new Abstract: Question answering (QA) is a core challenge in AI, particularly for complex queries requiring multi-hop reasoning across documents, or symbolic operations like aggregation or exhaustive listing.
Read next because Better Later Than Sooner: Neuro-Symbolic Knowledge Graph Construction via Ontology-grounded Post-extraction Correction overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, rect, correct, eval, token, line, extraction, stage. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29168v1 Announce Type: new Abstract: Question answering (QA) is a core challenge in AI, particularly for complex queries requiring multi-hop reasoning across documents, or symbolic operations like aggregation or exhaustive listing. Retrieval-augmented generation has become the dominant approach to QA, with recent graph-based variants addressing part of these issues by organizing knowledge to better support compositional questions. However, most textual graph-based RAG methods still lack the structure needed for symbolic operations useful to answer complex questions reliably. This motivates symbolic graph-based approaches, which extract knowledge graphs (KGs) whose relations are logic predicates that enable SQL-like querying. Yet these pipelines typically use LLMs for KG extraction, which can introduce consistency issues, where extracted facts may violate commonsense ontology constraints. We propose a neuro-symbolic framework for ontology-grounded KG construction combining open-domain extraction, embedding-based canonicalization of types and predicates, and targeted LLM-based correction of ontology violations. By deferring corrections to a post-extraction stage, our method avoids repeated LLM calls, substantially reducing token usage while improving KG consistency and preserving downstream QA quality. Finally, we show that the extracted KGs are well suited for symbolic querying by measuring the occurrence of SPARQL graph patterns.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29116unread
Beyond Consensus: Trace-Level Synthesis in Mixture of Agents
Shreyas Fadnavis, Praitayini Kanakaraj, Felix Wyss · 2026-05-29
arXiv:2605. 29116v1 Announce Type: new Abstract: When multiple LLM agents solve the same problem, standard practice compresses each agent's reasoning into a majority vote or layered synthesis, treating agreement as the finish line.
Read next because Beyond Consensus: Trace-Level Synthesis in Mixture of Agents overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: rect, correct, line, rate, does, chain, model, never. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29116v1 Announce Type: new Abstract: When multiple LLM agents solve the same problem, standard practice compresses each agent's reasoning into a majority vote or layered synthesis, treating agreement as the finish line. We show this is unnecessarily lossy: an LLM aggregator that reads complete reasoning traces recovers correct solutions even when agents unanimously agree, with beneficial corrections consistently outweighing harmful ones -- the \emph{aggregation paradox}. Majority voting has a ceiling that perturbation diversity does not raise (error correlations are identical); the aggregator's gain comes from trace-level complementarity, assembling correct intermediate steps from minority chains that voting discards. These findings motivate Self-Consistent Mixture of Agents which generates trace diversity through semantic-preserving input perturbations, safeguards the majority via anchored refinement with provable non-degradation guarantees, and always synthesizes -- never gates on consensus. A single model with perturbation-induced trace variation outperforms heterogeneous model pools across structured reasoning, PhD-level science, competition mathematics, and competitive programming. The unit of aggregation should be the reasoning trace, not the answer.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29096unread
Trends in AI and Human-AI Interaction in Clinical Trials -- A Hybrid Human-AI Exploration
Sandra Woolley, Tim Collins, Khalid Khattak, Illia Chernomorets, Ariane Arevalo, Chris Richardson · 2026-05-29
arXiv:2605. 29096v1 Announce Type: new Abstract: This paper examines records retrieved from the ClinicalTrials.
Read next because Trends in AI and Human-AI Interaction in Clinical Trials -- A Hybrid Human-AI Exploration overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)". Matching terms: class, good, screen, lora, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29096v1 Announce Type: new Abstract: This paper examines records retrieved from the ClinicalTrials.gov registry to characterize temporal trends in AI terminology and the geographical distribution of AI trials. The work also reports on an exploratory hybrid human-AI approach to analyzing human-AI interaction trends in registered clinical trials. The hybrid workflow comprised a frontier generative AI model (GPT-5.5) and human review to screen and categorize records returned by an AI-focused search. The findings indicate a marked increase in AI-related trials over time, with recent growth in references to machine learning, deep learning, chatbots, GPTs, and large language models. Geographically, China and the United States accounted for the largest numbers of AI-related trials, with notable recent increases in several other countries including Italy, France, Spain, the UK and Turkey (T\"urkiye). In a random sample of 100 records, human and AI classifiers showed good agreement in identifying studies not substantively using AI, but lower agreement in classifying human-AI interaction, particularly where health professional interaction was ambiguous or insufficiently described. Overall, the results suggest that hybrid human-AI screening of clinical trial records is potentially viable, but clearer trial reporting and more precise interaction definitions will benefit the process.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29042unread
Differentiable Belief-based Opponent Shaping
Aarav G Sane, Karthik Sivachandran, Rohan Paleja · 2026-05-29
arXiv:2605. 29042v1 Announce Type: new Abstract: Human coordination often relies on the ability to influence the beliefs of others through strategic action.
Read next because Differentiable Belief-based Opponent Shaping overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, soft, rate. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29042v1 Announce Type: new Abstract: Human coordination often relies on the ability to influence the beliefs of others through strategic action. In multi-agent reinforcement learning, opponent shaping attempts to replicate this influence, though existing methods typically operate within an opponent's parameter, policy, or value space. Meanwhile, belief-manipulation techniques in hidden-role games often rely on hard-coded objectives, such as deception or belief saturation. We propose Differentiable Belief-based Opponent Shaping (D-BOS), a first-order method that treats each observer's belief as the shaped opponent state and differentiates through $k$-step softmax-Bayes belief dynamics. Rather than explicitly rewarding deceptive or cooperative behavior, our method treats the belief state as the target for shaping. This allows the optimal strategy to emerge naturally from the environment's reward structure. This belief-space formulation provides an opponent-shaping signal by differentiating through opponent belief updates, and naturally extends to multiple observers by aggregating gradients over their individual inferred belief trajectories. Empirically, D-BOS outperforms PPO and BBM in hidden-role games, with the largest gains in mixed-motive settings.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29027unread
Mind Your Tone: Does Tone Alter LLM Performance?
Om Dobariya, Akhil Kumar · 2026-05-29
arXiv:2605. 29027v1 Announce Type: new Abstract: The use of Large Language Models (LLMs) is proliferating, yet their performance is observed to vary based on prompting styles and tones.
Read next because Mind Your Tone: Does Tone Alter LLM Performance? overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "Follow-up to #354: cascading chunk-binding — does A→B, B→C, C→D propagate the full chain on a recipient trained only to emit A?". Matching terms: eval, rate, does, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29027v1 Announce Type: new Abstract: The use of Large Language Models (LLMs) is proliferating, yet their performance is observed to vary based on prompting styles and tones. In this study, we investigate both whether and how tonal variations in prompts lead to disparate LLM accuracy for objective multiple-choice questions. We use two datasets: a 50-base question dataset with five tone variants and a 570-base question MMLU subset spanning 57 subjects with seven tone variants. Experiments were conducted to evaluate the performance of four cost-efficient, popular LLMs: ChatGPT-4o, ChatGPT-5-nano, Gemini 2.5 Flash, and Gemini 2.5 Flash Lite. Across models, tonal effects are systematic but highly model-dependent. Some models show small, yet statistically significant, shifts, while others exhibit large accuracy swings across tones. Further, we identify subject-level differences in tone sensitivity and present a routing framework to explain how tones may attune internal reasoning modes. Our findings caution users against assuming tone-robust reliability in LLM deployments.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.28897unread
Review Arcade: On the Human Alignment and Gameability of LLM Reviews
Hans Ole Hatzel, Sebastian Steindl, Jan Strich · 2026-05-29
arXiv:2605. 28897v1 Announce Type: new Abstract: LLM-generated reviews for scientific papers are gaining considerable traction and are even being officially piloted by major conferences.
Read next because Review Arcade: On the Human Alignment and Gameability of LLM Reviews overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, alignment, eval, rate, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.28897v1 Announce Type: new Abstract: LLM-generated reviews for scientific papers are gaining considerable traction and are even being officially piloted by major conferences. We have to assume that not only reviewers are using LLM-assistance, but also that authors use LLMs to revise their papers before submitting. In this work, we perform empirical experiments on papers from the 2025 ACL Rolling Review (ARR) to evaluate LLM reviews from both the author and the reviewer perspective. First, we identify a limited alignment of LLM reviews with human ones. In the best-case scenario, the alignment is reasonable. However, we also find that LLM-human alignment varies substantially across prompts and models. Finally, we investigate the scenario in which the author uses an iterative draft-revise workflow to improve the submission according to the LLM review. We find that this "gaming" of LLM reviews can be effective in specific scenarios, leading to a statistically significant increase of overall scores for up to 35\% of papers. We publish our code: https://github.com/uhh-hcds/reviewarcade.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.28883unread
Ultra-Reduced-Impact-Encased-Logging (URIEL): propose a new method for selective sustainable logging and post-harvest silvicultural treatment in tropical forest using airborne robotics systems
Daniel Albiero, Gelton Fernando de Morais, Daniela Han, Fl\'avio Roberto de Freitas Gon\c{c}alves, Artur Vit\'orio Andrade Santos, Wesllen Lins de Ara\'ujo, Alessandra Maia Freire, Cl\'audio Kiyoshi Umezu, Mateus Peressin, Francesco Toscano, Admilson \'Irio Ribeiro, Alfeu J. Sguarezi Filho, Am\'erico Ferraz Dias Neto, Angel Pontin Garcia · 2026-05-29
arXiv:2605. 28883v1 Announce Type: new Abstract: Tropical forests worldwide are under intense deforestation pressure driven by economic and political interests, and scientific evidence suggests this deforestation contributes to climate change.
Read next because Ultra-Reduced-Impact-Encased-Logging (URIEL): propose a new method for selective sustainable logging and post-harvest silvicultural treatment in tropical forest using airborne robotics systems overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, under, rate, factor. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.28883v1 Announce Type: new Abstract: Tropical forests worldwide are under intense deforestation pressure driven by economic and political interests, and scientific evidence suggests this deforestation contributes to climate change. This paper proposes a novel logging method for tropical forests, Ultra-Reduced-Impact-Encased-Logging (URIEL). This new method is based on heli-logging techniques combined with intensive use of robotics and AI integrated with post-harvest silvicultural treatments performed by drones. The concept of appropriate equipment for this method was developed, dimensions were determined, details were completed in a digital proof of concept, and an effective digital simulation and economic feasibility analysis were carried out for various helicopter-timber-distance combinations. The results demonstrated that a URIEL method has high economic viability and makes it possible to virtually eliminate collateral damage to forests while maintaining ecosystem services. The main conclusion of this paper is that, despite the satisfactory scientific and technological results, the feasibility of a Uriel method depends on the integration of stakeholders intrinsic to the context: high-tech industry; political governments; certified logging companies; and native populations.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27773unread
Do Models Know Why They Changed Their Mind? Interpretability and Faithfulness of Chain-of-Thought Under Knowledge Conflict
Pruthvinath Jeripity Venkata · 2026-05-29
arXiv:2605. 27773v1 Announce Type: new Abstract: When a language model sees a document contradicting its training knowledge, it must choose: follow the document or trust itself.
Read next because Do Models Know Why They Changed Their Mind? Interpretability and Faithfulness of Chain-of-Thought Under Knowledge Conflict overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, token, rate, sees, does, full, chain, test. Source: arxiv cs.CL (NLP).
arXiv:2605.27773v1 Announce Type: new Abstract: When a language model sees a document contradicting its training knowledge, it must choose: follow the document or trust itself. Prior work proved this choice depends on how well-known the fact is. We ask: does the model's chain-of-thought (CoT) reasoning faithfully report this mechanism? We introduce introspective faithfulness and test it across 200 questions, 8 models, and 4 prompt conditions. We find CoT reasoning is highly stable across opposite decisions: flip pairs retain 96% of same-answer similarity (d=0.34; confirmed by ROUGE-L, d=0.45). Yet self-rated confidence carries a faint genuine signal: for obscure facts where entity fame is uninformative, confidence still predicts decisions (p<0.001) and tracks item-level knowledge (r=0.134). GPT-4o is the only model with statistically reliable reasoning-decision coupling. Claude Sonnet 4.6 shows the widest confidence range (SD=1.39) but near-zero pooled correlation because the confidence-decision relationship reverses between conditions; a temperature ablation confirms this is model-specific. Internal thinking tokens show greater decision-sensitivity than user-facing CoT (p=0.033). CoT decomposes into a decision-invariant knowledge display (~96%) and a thin confidence layer with weak but real signal. For monitoring: read confidence, not the argument.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27642unread
Learning to Translate from Soft to Hard LLM Prompts
Pitipat Kongsomjit, Suryansh Goyal, Jacob Whitehill · 2026-05-29
arXiv:2605. 27642v1 Announce Type: new Abstract: Soft prompt tuning is a parameter-efficient method for adapting LLMs to specific tasks, but suffers from a lack of interpretability.
Read next because Learning to Translate from Soft to Hard LLM Prompts overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, soft, source, rate, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27642v1 Announce Type: new Abstract: Soft prompt tuning is a parameter-efficient method for adapting LLMs to specific tasks, but suffers from a lack of interpretability. Building on recent work on interpreting soft prompts (Ramati et al., 2024), we explore how training a dedicated soft prompt to natural language translation model can yield higher translation quality. In particular, in both quantitative and qualitative comparisons on multiple Datasets of Datasets (DoDs), we demonstrate that our translator produces fluent, accurate verbalizations that outperforms existing training-free methods like InSPEcT. In addition to advancing interpretability, our work suggests a promising downstream application: soft prompts optimized on small, open-source models can be translated into portable text prompts that, when deployed on larger closed-API models, exceed the performance of the original soft prompt and, in some cases, even few-shot learning.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27546unread
Keyphrase Generative Representation of Youth Crisis Conversations Beyond Static Taxonomies
Abeer Badawi, Will Aitken, Lydia Sequeira, Jocelyn Rankin, Maia Norman, Elham Dolatabadi · 2026-05-29
arXiv:2605. 27546v1 Announce Type: new Abstract: Crisis Responders (CRs) rapidly assess thousands of youth SMS conversations each year to identify mental health concerns and guide support.
Read next because Keyphrase Generative Representation of Youth Crisis Conversations Beyond Static Taxonomies overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, phrase, phrases, eval, rate, does, trained, language. Source: arxiv cs.CL (NLP).
arXiv:2605.27546v1 Announce Type: new Abstract: Crisis Responders (CRs) rapidly assess thousands of youth SMS conversations each year to identify mental health concerns and guide support. Yet youth distress is increasingly expressed through evolving and context-specific language that often does not fit fixed-label taxonomies. This work analyzed 703,975 de-identified Kids Help Phone conversations (2018-2023) and expanded KHP's 19-label issue taxonomy into a 39-label hierarchical schema. We then introduce Keyphrase Generative Representation (KGR), a constrained LLM generating concise, conversation-specific keyphrases, evaluated across 129 conversations and 387 expert annotations. The expanded taxonomy achieved expert consensus reliability, with an accuracy of 0.96, and expert review found that 81% of keyphrases accurately reflected content and 74% improved clarity. KGR surfaced identity-linked themes absent from the fixed taxonomy, including immigration problems and caregiver burden, and supported a topic-retrieval workflow that increased accuracy from 0.25 to 0.70 (+0.45) over the manual analyst process. KGR marks a shift toward hybrid, interpretable generative representations that extend crisis response beyond static taxonomies to surface emerging and culturally grounded patterns of youth distress.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27483unread
Debate Helps Weak Judges Reward Stronger Models
Ethan Elasky, Frank Nakasako, Naman Goyal · 2026-05-29
arXiv:2605. 27483v1 Announce Type: new Abstract: Despite theoretical promise, debate as a scalable oversight protocol has produced mixed empirical results: gains in some settings, and null effects in others, especially when the judge does not have information hidden from it.
Read next because Debate Helps Weak Judges Reward Stronger Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, strong, class, latin, line, rate, does, test. Source: arxiv cs.CL (NLP).
arXiv:2605.27483v1 Announce Type: new Abstract: Despite theoretical promise, debate as a scalable oversight protocol has produced mixed empirical results: gains in some settings, and null effects in others, especially when the judge does not have information hidden from it. We study proposer-critic debate in a stronger-debater/weaker-judge setting on programmatically verifiable code and logic tasks. Debate helps the judge over a consultancy baseline when the critic provides a usable advantage: the critic's classification ability must exceed the judge's, and the judge must treat critic speeches as claims to verify rather than testimony to summarize. On the three of five pairings where the condition holds, proposer-critic debate's gains are statistically significant over consultancy, and these pairings are the most capable model pairings. On the two non-responder pairings in our set, debate produces null effects, and judge verification rates drop by tens of percentage points once a critic enters the transcript. In these cases the critic's binary-classification ability and the judge's are within noise of each other, and the critic's disagreement is parsed as testimony rather than a claim to check. Ablating rebuttal rounds from debate produces no measurable change in judge performance: a single independent critique recovers the bulk of debate's benefit at lower inference cost. These findings suggest a cheaper primitive for training-free scalable oversight in verifiable domains (answer, critique, judge) and a pre-deployment audit (does the critic beat the judge, and will the judge verify it?) that predicts when debate will help.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27387unread
From AR to Diffusion: Efficiently Adapting Large Language Models with Strictly Causal and Elastic Horizons
Xiangyu Ma, Teng Xiao, Zuchao Li, Lefei Zhang · 2026-05-29
arXiv:2605. 27387v2 Announce Type: new Abstract: Diffusion models promise efficient parallel text generation but rely on bidirectional attention, creating a structural mismatch with pre-trained Autoregressive (AR) models.
Read next because From AR to Diffusion: Efficiently Adapting Large Language Models with Strictly Causal and Elastic Horizons overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, text, rect, alignment, rate, trained, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27387v2 Announce Type: new Abstract: Diffusion models promise efficient parallel text generation but rely on bidirectional attention, creating a structural mismatch with pre-trained Autoregressive (AR) models. This incompatibility precludes reusing robust AR priors, necessitating prohibitive pre-training from scratch. To bridge this gap, we propose FLUID, a framework that efficiently adapts AR backbones to the diffusion paradigm. By enforcing Strictly Causal Alignment, FLUID enables seamless initialization from standard GPT-style checkpoints, circumventing the need for massive pre-training. Furthermore, we introduce Elastic Horizons, an entropy-driven mechanism that dynamically modulates denoising strides based on local information density rather than fixed schedules. Experiments demonstrate that FLUID achieves state-of-the-art performance while reducing training costs by orders of magnitude, effectively reconciling established AR foundations with efficient parallel generation. Our code is available at https://github.com/Oli-lab-nun/FLUID/tree/main.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27383unread
Bridging the Stability-Expressivity Gap: Synthetic Data Scaling and Preference Alignment for Low-Resource Spoken Language Models
Yizhong Geng, Yanliang Li, Jinghan Yang, Tianhan Jiang, Boxun An, Ya Li, Xiaoyu Shen · 2026-05-29
arXiv:2605. 27383v1 Announce Type: new Abstract: Spoken Language Models (SLMs) have emerged as a promising paradigm for speech synthesis by bypassing explicit grapheme-to-phoneme pipelines.
Read next because Bridging the Stability-Expressivity Gap: Synthetic Data Scaling and Preference Alignment for Low-Resource Spoken Language Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, alignment, source, line, rate, capability, lora, language. Source: arxiv cs.CL (NLP).
arXiv:2605.27383v1 Announce Type: new Abstract: Spoken Language Models (SLMs) have emerged as a promising paradigm for speech synthesis by bypassing explicit grapheme-to-phoneme pipelines. However, their effectiveness in low-resource languages remains fundamentally limited by the scarcity of transcribed speech. In practice, synthetic data has become the primary strategy for scaling SLMs in such settings, providing reliable phonetic supervision when real data is insufficient. In this work, we show that this reliance introduces a fundamental trade-off, which we term the Stability-Expressivity Gap: while synthetic data improves phonetic accuracy, it progressively suppresses prosodic variability, ultimately leading to a collapse of expressivity (Synthetic Erosion). To bridge this gap, we propose two self-alignment frameworks. Disentanglement-Guided Self-Alignment (DGSA) recovers expressivity for complex languages by exploiting prosody-timbre separation. For regimes where authentic references are exceptionally limited, Temperature-Driven Self-Critique (TDSC) stabilizes generation through automated exploration and filtering. Our approach outperforms strong commercial systems, including ElevenLabs and Gemini Pro, and enables the first zero-shot voice cloning capability for Lao.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27377unread
RAG-Coding: Enhancing LLM Medical Coding with Structured External Knowledge
Yidong Gan, David D. Nguyen, Yang Lin, Peter Zhong, Thanh Vu, Long Duong, Yuan-Fang Li · 2026-05-29
arXiv:2605. 27377v2 Announce Type: new Abstract: Accurate medical coding requires consulting authoritative resources such as the ICD tabular list and coding guidelines.
Read next because Enhancing LLM Medical Coding with Structured External Knowledge overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, text, under, source, line, rate. Source: arxiv cs.CL (NLP).
arXiv:2605.27377v2 Announce Type: new Abstract: Accurate medical coding requires consulting authoritative resources such as the ICD tabular list and coding guidelines. Existing LLM-based automated methods largely rely on LLMs' internal knowledge, which is prone to hallucination and cannot keep pace with guideline updates. We introduce RAG-Coding, an agentic, training-free method that augments LLMs with structured external knowledge: the tabular list is encoded as a knowledge graph capturing hierarchical and instructional code relationships, and the guidelines are distilled into concise, code-specific summaries rather than retrieved as raw text. To enable our study, we also introduce MDACE-2025, expert re-annotations of the MDACE dataset under the 2025 ICD-10-CM/PCS guidelines, adding code sequencing and justification comments. On MDACE, RAG-Coding outperforms the best LLM-based baseline by 3--13\% in micro-F1 across five LLM backbones, and achieves comparable micro- and macro-F1 to the supervised state-of-the-art, with higher recall ($+$11\%) at the cost of precision ($-$6\%). On MDACE-2025, RAG-Coding outperforms all baselines, demonstrating effective generalisation to updated guidelines. Ablations confirm stepwise gains, highlighting the importance of integrating structured external knowledge for LLM-based medical coding.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27374unread
ICG: Improving Cover Image Generation via MLLM-based Prompting and Personalized Preference Alignment
Zhipeng Bian, Jieming Zhu, Qijiong Liu, Wang Lin, Guohao Cai, Zhaocheng Du, Jiacheng Sun, Zhou Zhao, Zhenhua Dong · 2026-05-29
arXiv:2605. 27374v1 Announce Type: new Abstract: Recent advances in multimodal large language models (MLLMs) and diffusion models (DMs) have opened new possibilities for AI-generated content.
Read next because ICG: Improving Cover Image Generation via MLLM-based Prompting and Personalized Preference Alignment overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, text, persona, title, under, alignment, token, line. Source: arxiv cs.CL (NLP).
arXiv:2605.27374v1 Announce Type: new Abstract: Recent advances in multimodal large language models (MLLMs) and diffusion models (DMs) have opened new possibilities for AI-generated content. Yet, personalized cover image generation remains underexplored, despite its critical role in boosting user engagement on digital platforms. We propose ICG, a novel framework that integrates MLLM-based prompting with personalized preference alignment to generate high-quality, contextually relevant covers. ICG extracts semantic features from item titles and reference images via meta tokens, refines them with user embeddings, and injects the resulting personalized context into the diffusion model. To address the lack of labeled supervision, we adopt a multi-reward learning strategy that combines public aesthetic and relevance rewards with a personalized preference model trained from user behavior. Unlike prior pipelines relying on handcrafted prompts and disjointed modules, ICG employs an adapter to bridge MLLMs and diffusion models for end-to-end training. Experiments demonstrate that ICG significantly improves image quality, semantic fidelity, and personalization, leading to stronger user appeal and offline recommendation accuracy in downstream tasks. As a plug-and-play adapter bridging MLLMs and diffusion models, ICG is compatible with common checkpoints and requires no ground-truth labels during optimization.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.29009unread
Label-Free Reinforcement Learning via Cross-Model Entropy
Matt Gorbett, Hossein Shirazi · 2026-05-29
arXiv:2605. 29009v1 Announce Type: new Abstract: Post-training large language models with reinforcement learning is bottlenecked by the reward signal.
Read next because Label-Free Reinforcement Learning via Cross-Model Entropy overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, rect, under, correct, eval, token, rate, trained. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.29009v1 Announce Type: new Abstract: Post-training large language models with reinforcement learning is bottlenecked by the reward signal. Existing approaches require either ground-truth verifiable rewards, restricting training to domains with automatic correctness checks (e.g., mathematics, code execution), or human preference labels, which are expensive to collect and prone to reward hacking. Recent label-free methods replace ground-truth verifiers with self-referential signals like majority voting or token entropy over a model's own outputs, but risk reinforcing a model's own errors. In this work we propose Cross-Model Entropy (CME), the mean log-likelihood of a generator's response under a separate verifier model, as a label-free reward signal for RL post-training. CME is continuous, training-free, and grounded in the principle that responses a verifier finds unsurprising are likely correct or high quality. Because the verifier is independent of the generator, the signal cannot be gamed through self-consistency. We integrate CME into GRPO with no other changes to the training loop, extending label-free RL to open-ended instruction following -- a regime where self-referential signals are inapplicable or poorly suited. On open-ended instruction following (UltraFeedback prompts, evaluated on AlpacaEval 2.0), CME rewards beat the untrained base in head-to-head LLM-as-Judge comparisons across four model families (Qwen, Llama, Gemma, OLMo) and three training regimes (pretrained, SFT, and instruction-tuned), with tie-adjusted win rates ranging from 52.5% to 71.4%. Code will be released upon publication.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28975unread
A Training-Time Diagnostic for Generalization via the Log-Alignment Ratio
Ali Shehper, Ashish Vaswani · 2026-05-29
arXiv:2605. 28975v1 Announce Type: new Abstract: We study the log-alignment ratio (LAR), a measure of parameter-activation alignment, introduced in parameterization theory.
Read next because A Training-Time Diagnostic for Generalization via the Log-Alignment Ratio overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, rect, alignment, line, rate, project, language, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28975v1 Announce Type: new Abstract: We study the log-alignment ratio (LAR), a measure of parameter-activation alignment, introduced in parameterization theory. We reformulate it as the overlap between a weight spectrum $p$ of the normalized squared singular values of a matrix and an activation spectrum $q$ of the normalized squared projections of inputs onto its singular directions. We show that unembedding LAR tracks the transition between memorization and generalization in two different settings by capturing the spread of $p$ and $q$ during training. In grokking, LAR predicts the effective dimension of the learned function: $k \approx n^{2(1-\text{LAR})}$, where $n$ is the input dimension of the matrix. In 3B-parameter language model pre-training, its deviation from a non-overfitting baseline tracks the generalization gap, and its rate of decline increases as overfitting approaches. LAR is computable from quantities available during the forward pass with negligible computational overhead, and requires no held-out validation data.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28920unread
Conf-Gen: Conformal Uncertainty Quantification for Generative Models
Gabriel Loaiza-Ganem, Kevin Zhang, Wei Cui, Marc T. Law, Kin Kwan Leung · 2026-05-29
arXiv:2605. 28920v1 Announce Type: cross Abstract: Conformal prediction (CP) and its extension, conformal risk control (CRC), are established frameworks for quantifying uncertainty in supervised machine learning through formal guarantees.
Read next because Conf-Gen: Conformal Uncertainty Quantification for Generative Models overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: rect, correct, rate, control, language, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.28920v1 Announce Type: cross Abstract: Conformal prediction (CP) and its extension, conformal risk control (CRC), are established frameworks for quantifying uncertainty in supervised machine learning through formal guarantees. However, recent breakthroughs in artificial intelligence (AI) have been driven by unsupervised generative models, such as large language models (LLMs) and image generators, which are not directly compatible with CP or CRC. In this work we introduce conformal generation (Conf-Gen), a general framework adapting CRC to generative tasks while relaxing its theoretical assumptions. Conf-Gen unifies and generalizes previous attempts to apply CP to LLMs, and extends conformal methodology to entirely new domains. We demonstrate the flexibility of Conf-Gen through some novel applications, including obtaining conformal guarantees on: image generators producing non-memorized images, conversational AI systems having asked enough clarifying questions, and the output of AI agents being correct.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28919unread
CosmicFish-HRM: Adaptive Reasoning via Hierarchical Recurrent Mechanisms in Compact Language Models
Venkat Akhil Lakkapragada · 2026-05-29
arXiv:2605. 28919v1 Announce Type: new Abstract: Large language models have achieved strong reasoning capabilities, though often at the cost of massive parameter counts and expensive inference.
Read next because CosmicFish-HRM: Adaptive Reasoning via Hierarchical Recurrent Mechanisms in Compact Language Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rect, rate, capability, language, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28919v1 Announce Type: new Abstract: Large language models have achieved strong reasoning capabilities, though often at the cost of massive parameter counts and expensive inference. In this work, we explore a different direction: adaptive reasoning depth in compact language models. We present CosmicFish-HRM, a compact language model built around a Hierarchical Reasoning Module (HRM) that dynamically allocates computational effort during inference. Instead of applying fixed computation to every input, the model iterates through high-level and low-level reasoning cycles and learns when to halt based on input complexity. CosmicFish-HRM combines this adaptive reasoning core with modern transformer components including Grouped Query Attention, RoPE, and SwiGLU activations. While the additional reasoning infrastructure introduces overhead at small scale, we hypothesize that this tradeoff becomes increasingly favorable as model size grows and the relative cost of the HRM core diminishes. Our results show that the model learns non-uniform reasoning behavior, allocating different numbers of reasoning steps across tasks and inputs. These findings suggest that adaptive reasoning depth may offer a promising alternative to relying solely on parameter scale for reasoning capability.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28900unread
Spectral Guidance for Flexible and Efficient Control of Diffusion Models
Gabriel Moreira, Manuel Marques, Jo\~ao Paulo Costeira, Chenyan Xiong · 2026-05-29
arXiv:2605. 28900v1 Announce Type: new Abstract: We introduce Spectral Guidance, a framework for controlling diffusion models by leveraging the intrinsic geometry of the generative process.
Read next because Spectral Guidance for Flexible and Efficient Control of Diffusion Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rect, line, project, control, without, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28900v1 Announce Type: new Abstract: We introduce Spectral Guidance, a framework for controlling diffusion models by leveraging the intrinsic geometry of the generative process. As data is progressively corrupted by noise, only a small number of features remain informative for control. We characterize them as the singular functions of a conditional expectation operator and show that they can be learned via a self-supervised objective. Once recovered, this basis enables the projection of arbitrary guidance signals, such as labels, CLIP embeddings, or masks, directly onto the sampling trajectory. This approach allows for stable, high-fidelity control without retraining or denoiser backpropagation during sampling. Empirically, we improve conditional accuracy on CIFAR-10 by 37 percentage points over the strongest training-free baseline while offering $4\times$ faster sampling. Moreover, the same representations that support label and CLIP guidance also enable spatial control, such as mask-based guidance, without auxiliary models. Finally, our framework reveals a phase transition in the generative process, pinpointing the optimal time window for effective guidance.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28896unread
Feature Geometry of LoRA Adapters: A Sparse Autoencoder Analysis of Representational Divergence in Fine-Tuned Language Models
Prasanth K K · 2026-05-29
arXiv:2605. 28896v1 Announce Type: new Abstract: Low-Rank Adaptation (LoRA) has emerged as a widely adopted approach for adapting large language models, yet the internal representational changes induced by LoRA fine-tuning remain insufficiently understood.
Read next because Feature Geometry of LoRA Adapters: A Sparse Autoencoder Analysis of Representational Divergence in Fine-Tuned Language Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, rect, under, alignment, eval, compare, full, trained. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28896v1 Announce Type: new Abstract: Low-Rank Adaptation (LoRA) has emerged as a widely adopted approach for adapting large language models, yet the internal representational changes induced by LoRA fine-tuning remain insufficiently understood. In this work, we investigate the geometry of LoRA-induced representations using Sparse Autoencoders (SAEs). We introduce a delta activation framework that isolates the adapter-specific contribution to the residual stream. Using Gemma-2-9B with LoRA ranks 4, 8, 16, and 32, we train adapter-specific SAEs across multiple transformer layers and compare their learned feature spaces with pretrained SAE dictionaries. We evaluate representational alignment using cosine similarity between decoder directions, principal-angle analysis of feature subspaces, and Centered Kernel Alignment (CKA) between activation representations. Across layers and ranks, we consistently observe comparatively weak geometric alignment between LoRA-induced feature dictionaries and pretrained SAE features. Adapter-specific SAEs also reconstruct delta activations more effectively than pretrained SAEs, suggesting that LoRA updates occupy partially distinct representational structure within the residual stream. Additionally, feature density increases with rank and depth, while geometric divergence remains relatively stable across ranks. These findings provide empirical evidence that LoRA fine-tuning can induce feature structures that are not fully captured by pretrained interpretability dictionaries, with implications for mechanistic interpretability, adaptation analysis, and safety auditing of fine-tuned language models.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28880unread
Towards Continuous-time Causal Foundation Models
Dennis Thumm, Ruben Wiedemann, Ying Chen · 2026-05-29
arXiv:2605. 28880v1 Announce Type: new Abstract: Extending discrete-time causal Prior-data Fitted Networks for time series to continuous time invites writing the mechanism as a stochastic differential equation (SDE) -- but if the SDE is integrated \emph{once per observation gap}, the trajectory law depends on when it is observed, and the prior remains a discrete-time Markov model in SDE clothing.
Read next because Towards Continuous-time Causal Foundation Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, soft, eval, line, rate, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28880v1 Announce Type: new Abstract: Extending discrete-time causal Prior-data Fitted Networks for time series to continuous time invites writing the mechanism as a stochastic differential equation (SDE) -- but if the SDE is integrated \emph{once per observation gap}, the trajectory law depends on when it is observed, and the prior remains a discrete-time Markov model in SDE clothing. We propose a precise continuity criterion -- trajectory-law invariance to the observation schedule -- together with a three-tier taxonomy (discrete; naive observation-grid integration; fine-grid integration with decoupled observation) and a construction realising the top tier on a random DAG with OU or small-MLP nonlinear drifts, irregular observation schedules, and hard / soft / time-varying interventions. A $2 \times 2$ encoder $\times$ integrator ablation, run independently on a linear and a nonlinear prior, finds fine-grid integration beats naive on 8/8 cells (sign-consistency $p < 1/256$) with the gap growing as the eval grid refines; the encoder axis is null with fine integration but time-aware-leading with naive. We release the prior and a preliminary zero-shot protocol on pharmacokinetic and physical-system data.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28869unread
Balancing Multimodal Learning through Label Space Reshaping
Xiaoyu Ma, Weijie Zhang, Yuanhao Gao, Han Miao, Yongjian Deng, Hao Chen · 2026-05-29
arXiv:2605. 28869v1 Announce Type: new Abstract: Multimodal learning often suffers from modality imbalance, where modalities that converge faster dominate optimization while others remain undertrained.
Read next because Balancing Multimodal Learning through Label Space Reshaping overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, strong, class, under, source, rate, without, trained. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28869v1 Announce Type: new Abstract: Multimodal learning often suffers from modality imbalance, where modalities that converge faster dominate optimization while others remain undertrained. Existing approaches typically mitigate this issue by strengthening the weak modality or adjusting optimization gradients. However, such strategies mainly compensate for optimization rate discrepancies, often at the expense of the strong modality's optimization capacity, without analyzing how these discrepancies arise at the modality level. Based on theoretical insights and empirical observations, we argue that the discrepancy of learning pace arises from differences in the mapping difficulty between modality-specific feature space and the shared label space. To address this issue, we propose Balanced Multimodal Label Reshaping (BMLR), the first method that promotes multimodal balance from the label-side design. BMLR reshapes the cross-modal label space to equalize mapping difficulty across modalities, thereby facilitating modality interaction and injecting richer inter-class information into each modality. Extensive experiments across multiple architectures demonstrate that BMLR consistently improves multimodal performance and exhibits strong compatibility with diverse model designs. The source code will be released soon.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28868unread
TaxDistill: Improving Metagenomic Taxonomic Annotation via Distilled Genomic Foundation Models
Rongye Ye, Lun Li, Zheng Luo, Yiran Zhan, Shuhui Song · 2026-05-29
arXiv:2605. 28868v1 Announce Type: new Abstract: Metagenomic taxonomic annotation aims to identify the microbial origins of DNA fragments in environmental samples.
Read next because TaxDistill: Improving Metagenomic Taxonomic Annotation via Distilled Genomic Foundation Models overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)". Matching terms: class, rect, correct, soft, eval, line, rate, trained. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28868v1 Announce Type: new Abstract: Metagenomic taxonomic annotation aims to identify the microbial origins of DNA fragments in environmental samples. Traditional methods that rely on sequence similarity are often constrained by the high microbial diversity and the incompleteness of reference databases, which has motivated the development of learning approaches such as Taxometer that perform post hoc correction to learn more informative metagenomic sequence representations. However, these methods typically rely on labels derived from similarity search tools during training, which inevitably introduces noise that can impair representation learning and degrade classification performance. To address this issue, we propose TaxDistill, a knowledge distillation framework for metagenomic classification. We introduce GenomeOcean, a 500M parameter genomic foundation model, as the teacher network to extract deep semantic features and generate soft labels based on confidence. By distilling this soft label information into a lightweight student network, TaxDistill effectively reduces the label noise introduced by initial retrieval tools. Comprehensive experiments on seven diverse CAMI2 datasets demonstrate that TaxDistill outperforms existing baselines in most scenarios. For instance, on the Gastrointestinal dataset, it improves the F1 score of MMseqs2 from 0.763 to 0.941, outperforming the Taxometer baseline. Overall, TaxDistill provides a reliable method for label correction in complex metagenomic analysis.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28860unread
Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?
Jeanmely Rojas Nunez, Viraj Sawant, Nathan Allen, Nomgondalai Amgalanbaatar, Yannis Zongo, Vasu Sharma, Maheep Chaudhary · 2026-05-29
arXiv:2605. 28860v1 Announce Type: new Abstract: Fine-tuning large language models (LLMs) frequently induces catastrophic forgetting of prior capabilities.
Read next because Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT? overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, strong, under, compare, qwen2, language, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28860v1 Announce Type: new Abstract: Fine-tuning large language models (LLMs) frequently induces catastrophic forgetting of prior capabilities. Recent work has shown that reinforcement learning (RL) retains prior capabilities more effectively than supervised fine-tuning (SFT), attributing this to policy-gradient updates remaining closer to the base policy \cite{shenfeld2025rl}. We extend this behavioral account to the mechanistic level and ask whether RL's advantage is mirrored by stronger preservation of internal computational circuits. We introduce differential circuit vulnerability, a head-level measure of how much a circuit degrades under fine-tuning, and use it to compare RL and SFT on Qwen2.5-3B-Instruct adapted to scientific question-answering. We find a clear mechanistic trade-off: SFT adapts more rapidly to the target task but produces substantially greater circuit disruption and forgetting of prior capabilities, whereas RL preserves a larger fraction of the base circuit at the cost of slower task adaptation. These findings suggest that circuit preservation may help explain why RL is more robust to catastrophic forgetting. We released our code here: https://github.com/rl-sft-circuit-research/differential-circuit-vulnerability.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28839unread
One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them
Ali Holmov, Paul Youssef, Nandi Schoots, Christin Seifert · 2026-05-29
arXiv:2605. 28839v1 Announce Type: new Abstract: Knowledge editing methods such as ROME and MEMIT update factual associations in transformer models by modifying MLP weights.
Read next because One Mask to Rule Them All: On Hidden Facts after Editing and How to Find Them overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", experiment "Follow-up to #354: cascading chunk-binding — does A→B, B→C, C→D propagate the full chain on a recipient trained only to emit A?". Matching terms: under, eval, propagate, emit, test, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28839v1 Announce Type: new Abstract: Knowledge editing methods such as ROME and MEMIT update factual associations in transformer models by modifying MLP weights. While evaluated mainly by output behavior, their internal mechanism remains underexplored. We investigate whether edits rely on a common mechanism, regardless of which fact is modified. Despite fact-specific weight changes, we argue that ROME and MEMIT target the same subset of weights critical for maintaining edits. To isolate this subset, we train a compact binary mask over the edited weights. The mask reverses 80% of edits on the training set and over 70% on the test set, confirming that diverse edits share a common functional structure. Our analysis reveals that the mask reverses edits by eliminating overattention in later layers. Additionally, we show that injecting the mask during editing drops editing success from 98% to 38%, demonstrating that this mechanism is necessary for edits to succeed. Our finding that edits suppress rather than overwrite knowledge explains why ROME and MEMIT fail to propagate changes to related facts. The identified common functional subspace informs detection and defense against unwanted edits.
source ↗open
score 94arxiv cs.CR (Cryptography and Security)arxiv:2605.28991unread
A Secure, Manifest-Based Framework for Delegated Privilege Promotion
Rajarshi Chowdhury, Akshay Shah · 2026-05-29
arXiv:2605. 28991v1 Announce Type: new Abstract: Large-scale enterprise software systems commonly run as unprivileged service accounts to enforce least privilege, yet still depend on a small set of privileged components -- such as executables with elevated ownership, permissions, or capabilities -- for narrowly scoped operations.
Read next because A Secure, Manifest-Based Framework for Delegated Privilege Promotion overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone". Matching terms: soft, line, without, full. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.28991v1 Announce Type: new Abstract: Large-scale enterprise software systems commonly run as unprivileged service accounts to enforce least privilege, yet still depend on a small set of privileged components -- such as executables with elevated ownership, permissions, or capabilities -- for narrowly scoped operations. This creates a persistent security and operational conflict during maintenance. Automated patching tools running without elevated privileges cannot safely update privileged components without either executing the entire patch with full administrative rights or requiring manual administrator intervention. We present a secure, manifest-based infrastructure for delegated promotion of privileged software components, deployed in production as part of a large-scale enterprise database system serving both cloud and on-premises installations. The design centers on a minimal privileged mediator that validates cryptographically protected metadata and allows an unprivileged process to promote only vendor-approved files. The system explicitly mitigates Time-of-Check-to-Time-of-Use (TOCTOU) attacks using file-descriptor-bound validation and promotion, supports offline key rotation and revocation, and enables zero-downtime self-update via atomic replacement.
source ↗open
score 94arxiv cs.AI (Artificial Intelligence)arxiv:2605.29129unread
Governing Technical Debt in Agentic AI Systems
Muhammad Zia Hydari, Raja Iqbal, Narayan Ramasubbu · 2026-05-29
arXiv:2605. 29129v1 Announce Type: new Abstract: Agentic AI systems are increasingly being explored as production infrastructure: they reason over multiple steps, call tools, act through workflows, and adapt through memory and feedback.
Read next because Governing Technical Debt in Agentic AI Systems overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone". Matching terms: soft, line, control, full. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29129v1 Announce Type: new Abstract: Agentic AI systems are increasingly being explored as production infrastructure: they reason over multiple steps, call tools, act through workflows, and adapt through memory and feedback. These systems create governance challenges that are not fully captured by traditional software or predictive ML technical debt. We define Agentic Technical Debt as the accumulated liability created when prompts, memory, tool schemas, orchestration graphs, control policies, and observability routines are patched together faster than they can be validated, standardized, and governed. We define Stochastic Tax as the recurring operating burden of keeping probabilistic agent behavior within acceptable bounds. The distinction matters: debt is a stock of design and governance liability, while the tax is a flow of operating cost that arises because stochastic agents act through tools and workflows. We outline how managers can make both visible through lightweight dashboards and governance controls.
source ↗open
score 82arxiv cs.CR (Cryptography and Security)arxiv:2605.29465unread
Bridging Theory and Practice: An Executable Taxonomy of Security Properties for ProVerif and Tamarin
Leonard Tudorache, Ivan Kurtev, Mark van den Brand · 2026-05-29
arXiv:2605. 29465v1 Announce Type: new Abstract: Security is critical for everything relying on modern digital systems.
Read next because Bridging Theory and Practice: An Executable Taxonomy of Security Properties for ProVerif and Tamarin overlaps with clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)", experiment "Implement Chen et al. persona-vector extraction recipe and compare to project's centroid-difference recipe", experiment "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: source, implement, language, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29465v1 Announce Type: new Abstract: Security is critical for everything relying on modern digital systems. Because almost all digital interactions are governed by the Internet and cryptographic protocols, these protocols must serve as reliable mechanisms that guarantee core security properties, such as confidentiality and integrity. Formal verification of these protocols is a critical step in securing interconnected systems. Tools such as ProVerif and Tamarin are widely employed to perform automated verification. However, their effective use demands specialized domain knowledge, creating a significant learning curve for security protocol designers who often have a security, rather than a formal verification background. We therefore need structured, accessible resources to help protocol designers to express their design and requirements in the language of the formal verification tools. To address this, we introduce a systematic and evidence-based taxonomy of security properties. This taxonomy is derived from a literature review of 53 recent studies (2022-2025) that used ProVerif and Tamarin, providing an up-to-date view of verified properties. We systematically categorize and define these properties, providing both informal definitions for intuitive comprehension and rigorous formal definitions expressed in first-order logic for clarity and consistency. We further detail modeling patterns and implement executable examples in both ProVerif and Tamarin, collected in an open repository. This work advances the state of the art by bridging the gap between theoretical security property definitions and their practical, executable verification models.
source ↗open
score 78arxiv stat.ML (Machine Learning)arxiv:2605.29908unread
Joint Model and Data Sparsification via the Marginal Likelihood
Alexander Timans, Thomas M\"ollenhoff, Christian A. Naesseth, Mohammad Emtiyaz Khan, Eric Nalisnick · 2026-05-29
arXiv:2605. 29908v1 Announce Type: new Abstract: Sparse recovery in linear systems underpins applications from signal processing to high-dimensional regression.
Read next because Joint Model and Data Sparsification via the Marginal Likelihood overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: under, line, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29908v1 Announce Type: new Abstract: Sparse recovery in linear systems underpins applications from signal processing to high-dimensional regression. Sparse Bayesian Learning, grounded in the principle of automatic relevance determination (ARD), offers a practical Bayesian mechanism for feature sparsity via marginal likelihood optimization. Yet, its reliance on a homoscedastic noise model renders it sensitive to data contaminations such as outliers or misspecified noise, harming model fit and predictions. Instead, we propose jointly learning individual feature and sample relevancies, enabling simultaneous model and data sparsification via a single Bayesian objective. This symmetric pruning of model and data offers a natural extension that preserves conjugacy, admits closed-form updates for standard optimization procedures, and aligns with perspectives from robust regression and influence functions. Empirical results across diverse regression tasks affirm that a joint ARD approach consistently yields both sparse and robust prediction models.
source ↗open
score 78arxiv cs.CR (Cryptography and Security)arxiv:2605.29169unread
Domain-Informed Representation for Evolutionary Sieving in Integral and Module Lattices
Ahmad Tashfeen, Qi Cheng · 2026-05-29
arXiv:2605. 29169v1 Announce Type: new Abstract: Traditional cryptography, rooted in problems, e.
Read next because Domain-Informed Representation for Evolutionary Sieving in Integral and Module Lattices overlaps with experiment "Follow-up to #354: cascading chunk-binding — does A→B, B→C, C→D propagate the full chain on a recipient trained only to emit A?", experiment "Factor screen for marker implantation + leakage (2^5: system-prompt length, answer-format length, persona-presence, on-policy, marker-only-loss)", experiment "Test FR↔IT bystander-spill symmetry at multi-seed + 5 phrasings — pooled-rate vs per-phrasing asymmetry from #239 fact-check". Matching terms: full, factor, test. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29169v1 Announce Type: new Abstract: Traditional cryptography, rooted in problems, e.g., integer factorisation or discrete log, is inevitably vulnerable to a fully operational quantum computer. Although it remains an engineering frontier, the looming threat extends to encrypted data stored today, which could be decrypted in the future with quantum capabilities. To safeguard against this eventuality, the backbone of the modern quantum-safe cryptography is the Shortest Vector Problem (SVP). We enhance Laarhoven's treatment of Ajtai et al.'s sieving as a genetic algorithm (GA) for the SVP by incorporating domain-informed SVP representation and crossover while naturally extending application to the module lattices.
source ↗open
score 74arxiv cs.CR (Cryptography and Security)arxiv:2605.28952unread
Optimal Rates for Differentially Private Hypothesis Testing with E-values
Ben Jacobsen, Tomas Gonzales, Gavin Brown, Kassem Fawaz, Aaditya Ramdas · 2026-05-29
arXiv:2605. 28952v1 Announce Type: new Abstract: E-values have attracted considerable interest in recent years as flexible tools for enabling anytime-valid and adaptive data analysis.
Read next because Optimal Rates for Differentially Private Hypothesis Testing with E-values overlaps with clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "#351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation", experiment "Test FR↔IT bystander-spill symmetry at multi-seed + 5 phrasings — pooled-rate vs per-phrasing asymmetry from #239 fact-check". Matching terms: rate, test. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.28952v1 Announce Type: new Abstract: E-values have attracted considerable interest in recent years as flexible tools for enabling anytime-valid and adaptive data analysis. Hypothesis testing is at the core of many of these applications, which can often involve private or sensitive data. In this work, we answer a simple but important question: given two distributions $\mathbb{P}$ and $\mathbb{Q}$, what is the maximum achievable e-power when testing $X\sim \mathbb{P}^n$ against $X\sim\mathbb{Q}^n$ with e-values that satisfy $\varepsilon$-differential privacy? We characterize the optimal rate for this problem and provide an algorithm which matches it exactly. In the sequential setting, when observations arrive one-by-one and the analyst chooses when to halt, we give matching upper and lower bounds on the stopping times of any private e-process. Numerical experiments confirm the practicality of our algorithms, which require less data than the recently proposed DP-SPRT across a range of sequential testing problems and privacy levels.
source ↗open
score 62arxiv stat.ML (Machine Learning)arxiv:2605.30253unread
Wasserstein Contraction of Coordinate Ascent Variational Inference
Rocco Caprio, Adrien Corenflos, Sam Power · 2026-05-29
arXiv:2605. 30253v1 Announce Type: new Abstract: We study the contraction in Wasserstein distance of the coordinate ascent variational inference algorithm.
Read next because Wasserstein Contraction of Coordinate Ascent Variational Inference overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", experiment "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: under, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.30253v1 Announce Type: new Abstract: We study the contraction in Wasserstein distance of the coordinate ascent variational inference algorithm. This is shown to hold under a transport-information inequality at the fixed points and a functional smoothness condition. The results are general and sharp, allow for local convergence guarantees, hold for general smooth manifolds, and also in some non-smooth spaces. We consider applications to Bayesian Gaussian Mixture Models, and high-dimensional Bayesian Probit Regression, and Logistic Regression with P\'olya-Gamma random variables (i.e. Jaakkola-Jordan's algorithm).
source ↗open
score 46arxiv stat.ML (Machine Learning)arxiv:2605.29415unread
Constructing efficient channels for ideal observers using the conjugate gradient method
Weimin Zhou · 2026-05-29
arXiv:2605. 29415v1 Announce Type: cross Abstract: Task-based assessment of image quality (IQ) is critically important for the design and optimization of medical imaging systems.
Read next because Constructing efficient channels for ideal observers using the conjugate gradient method overlaps with clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: line. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29415v1 Announce Type: cross Abstract: Task-based assessment of image quality (IQ) is critically important for the design and optimization of medical imaging systems. Ideal observers, including the Bayesian Ideal Observer (IO) and the ideal linear observer, i.e., the Hotelling observer (HO), provide objective figures of merit (FOMs) that quantify system performance on signal detection tasks. However, the application of ideal observers to high-dimensional image data is often computationally intractable. Channel mechanisms provide an effective framework for dimensionality reduction that can facilitate the computation of ideal observers. This work presents a conjugate gradient (CG)-based method to construct efficient channels for approximating the IO and HO performance.
source ↗open

Threats and caveats

score 100arxiv stat.ML (Machine Learning)arxiv:2605.29836unread
CB-SLICE: Concept-Based Interpretable Error Slice Discovery
Yael Konforti, Mateo Espinosa Zarlenga, Elaf Almahmoud, Mateja Jamnik · 2026-05-29
arXiv:2605. 29836v1 Announce Type: cross Abstract: Despite strong average-case performance, deep learning models often exhibit systematic errors on specific population groups, known as error slices.
Read next because CB-SLICE: Concept-Based Interpretable Error Slice Discovery overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, word, rect, under, source, rate, candidate, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29836v1 Announce Type: cross Abstract: Despite strong average-case performance, deep learning models often exhibit systematic errors on specific population groups, known as error slices. Identifying these groups and the root causes of their failures is critical for model debugging and bias mitigation. However, existing error Slice Discovery Methods (SDMs) typically generate explanations disconnected from the model's inference process, thus only approximating the underlying error source and may be inaccurate. We address this limitation by leveraging Concept Bottleneck Models (CBMs), whose predictions are directly dependent on human-understandable semantic concepts. Since downstream task failures in CBMs commonly arise from concept mispredictions, concept representations provide a strong candidate for error slice identification, offering fine-grained explanations directly linked to the error source. Building on this insight, we introduce CB-SLICE, a concept-based SDM that groups samples with shared concept prediction failures and identifies the keyword concepts most responsible for each slice's failure mode. Across multiple benchmarks, we show that CB-SLICE outperforms state-of-the-art methods in uncovering well-known biases while providing richer and more faithful explanations of model errors.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures, limitation, bias, benchmark.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29684unread
Kernel Renormalization in Bayesian Deep Neural Networks: the Equivalent Wishart Ansatz in the Proportional Regime
Paolo Baglioni, Christian Keup, Vincenzo Zimbardo, Rosalba Pacelli, Alessandro Vezzani, Raffaella Burioni, Pietro Rotondo · 2026-05-29
arXiv:2605. 29684v1 Announce Type: cross Abstract: The scaling limit where both the size of the training set $P$ and the width $N$ of a deep neural network grow at the same rate, the so-called proportional-width regime, has been intensely studied for shallow, single-hidden-layer networks.
Read next because Kernel Renormalization in Bayesian Deep Neural Networks: the Equivalent Wishart Ansatz in the Proportional Regime overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, strong, class, width, good, line, rate, test. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29684v1 Announce Type: cross Abstract: The scaling limit where both the size of the training set $P$ and the width $N$ of a deep neural network grow at the same rate, the so-called proportional-width regime, has been intensely studied for shallow, single-hidden-layer networks. However, extending these non-perturbative results from shallow architectures to deep non-linear networks has proven very challenging. Here we present an effective approximate approach to predict the generalization performance of Bayesian multi-layer perceptrons (MLPs) of fixed depth $L$ on arbitrary high-dimensional data. We propose an equivalent Wishart Ansatz to capture the dominant stochastic fluctuations of the hierarchical empirical kernels of MLPs. This allows us to perform a large deviation analysis for the partition function of MLPs in the proportional limit, expressed in terms of a renormalized NNGP kernel. In this description, even strong representation learning in the proportional limit is encoded in at most $L$ scalar order parameters, determined self-consistently. Extending the approach to convolutional architectures (CNNs), we identify a hierarchical local kernel renormalization mechanism, which allows to quantify more complex data-dependent transformations of the large-width kernel in CNNs due to finite-width effects. We test our effective theory against sampling experiments from the Bayesian posterior of finite deep neural networks with depths $L \sim O(10)$ and $P\sim O(10^3)$ on classic benchmark datasets, finding overall very good agreement together with two distinct types of systematic deviations.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29580unread
On the Construction and Implications of Low-Loss Valleys in LoRA-based Bayesian Inference
Daniel Dold, Emanuel Sommer, Julius Kobialka, Oliver D\"urr, David R\"ugamer · 2026-05-29
arXiv:2605. 29580v1 Announce Type: cross Abstract: While parameter-efficient fine-tuning methods like low-rank adaptation (LoRA) are standard for large language models, principled estimation of epistemic uncertainty remains challenging.
Read next because On the Construction and Implications of Low-Loss Valleys in LoRA-based Bayesian Inference overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)". Matching terms: class, line, control, without, lora, qwen2, language, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29580v1 Announce Type: cross Abstract: While parameter-efficient fine-tuning methods like low-rank adaptation (LoRA) are standard for large language models, principled estimation of epistemic uncertainty remains challenging. Recent results in the LoRA regime suggest that discrete multi-mode approaches such as deep ensembles offer little benefit over single-mode methods. This contradicts broader observations in deep learning, where ensembling independent optima typically improves generalization, and linking these modes through continuous low-loss valleys further enhances Bayesian model averaging (BMA). Whether such structure exists in the LoRA space and whether it yields functional diversity missed by local or discrete methods has not been studied. We introduce LoRA-Curve, a segmented B\'ezier curve parameterization in the LoRA space, with two variants: a free configuration that jointly optimizes all control points, and an anchored configuration that connects independently fine-tuned LoRA optima. We prove pathwise continuity and Lipschitz regularity of the loss along the curve and empirically show, across reasoning and classification benchmarks with Qwen2.5 7B, that linear interpolation encounters loss barriers, while our anchored multi-segment curves connect independent optima through continuous low-loss valleys. Combined with flat-minima perturbations and a Jensen-Shannon divergence regularizer, LoRA-Curve yields measurably higher mutual information of the predictive distribution without sacrificing performance, and links continuous parameter-space traversal to functional diversity.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29411unread
The Good, the Bad, and the Ugly of Markov Boundary for Tabular Prediction
Shu Wan, Abhinav Gorantla, Huan Liu, K. Sel\c{c}uk Candan · 2026-05-29
arXiv:2605. 29411v1 Announce Type: cross Abstract: Under standard graphical assumptions, the Markov boundary of a target variable is the smallest set of features that renders every other feature redundant.
Read next because The Good, the Bad, and the Ugly of Markov Boundary for Tabular Prediction overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: under, good, eval, line, does, full, trained, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29411v1 Announce Type: cross Abstract: Under standard graphical assumptions, the Markov boundary of a target variable is the smallest set of features that renders every other feature redundant. Once the boundary is observed, the target is conditionally independent of the rest of the table. This is a tempting object for tabular prediction, since it names exactly the columns a model should need. Yet modern regressors are still trained on the full feature set. We ask whether the Markov boundary is genuinely useful for prediction on SCM3K, a 3,450-task synthetic SCM benchmark with feature counts from 40 to 1000 and six SCM families, evaluated with six regressors. The answer is more nuanced than the theory suggests. Restricting a regressor to the oracle boundary often improves prediction substantially, and the improvement grows as the feature space becomes larger and sparser. But the natural pipeline of recovering the boundary with causal discovery and training on the recovered mask does not deliver. Existing estimators exhaust the compute budget before reaching the regime where the boundary helps most, and even where they run they rarely beat the full feature set. We trace this to three causes. Discovery optimizes structural recovery rather than prediction. False negatives and false positives carry sharply asymmetric predictive cost. The exact boundary is only one of many feature sets that beat all features. We then develop what these facts imply for prediction-aligned feature selection and for tabular models that learn to use causal structure.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses negative, benchmark.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29395unread
Low Rank for Rank: Uncertainty-Aware Task-Specific LLM Ranking under Sparse Pairwise Comparisons
Jiachun Li, David Simchi-Levi, Will Wei Sun · 2026-05-29
arXiv:2605. 29395v1 Announce Type: cross Abstract: Pairwise human-preference platforms such as Chatbot Arena have become central to large language model (LLM) evaluation, yet reliable task-specific ranking remains challenging.
Read next because Low Rank for Rank: Uncertainty-Aware Task-Specific LLM Ranking under Sparse Pairwise Comparisons overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, eval, rate, test, language, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29395v1 Announce Type: cross Abstract: Pairwise human-preference platforms such as Chatbot Arena have become central to large language model (LLM) evaluation, yet reliable task-specific ranking remains challenging. Global leaderboards mask task heterogeneity, while ranking each fine-grained task independently is unstable under sparse, imbalanced comparisons. We propose a low-rank framework for task-specific LLM ranking from sparse pairwise comparisons, modeling the task-by-model ability matrix $\Theta^\star \in \mathbb{R}^{d_t \times d_m}$ as low rank so that information is shared across related tasks while task-specific differences are preserved. We first develop a max-norm ($\ell_\infty$) accurate estimator for the latent scores, combining a convex initializer with alternating-minimization refinement, and prove task-wise top-$K$ recovery guarantees under sparse sampling. Our main contribution is an uncertainty quantification framework for task-specific ranking. We construct cross-fitted one-step debiased estimators for fixed score contrasts -- such as the task-specific ability gap between two models -- yielding asymptotically valid confidence intervals that attain the semiparametric efficiency bound. We then extend the inference to the high-dimensional ranking regime, where per-task ranks and top-$K$ membership are determined by many dependent score-gap hypotheses. Using Gaussian and multiplier-bootstrap calibration, we obtain simultaneous confidence sets for per-task ranks and valid top-$K$ membership tests across many tasks and models. Experiments on synthetic data and Chatbot Arena show that low-rank sharing improves sample efficiency over independent task-wise Bradley-Terry estimation and produces tighter, better-calibrated ranking certificates, with the largest gains in the sparse regime typical of real LLM benchmarks.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses bias, evaluation, benchmark.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29371unread
Kernel-based potential mean-field games with unbiased random Fourier $U$-statistics
Yumiharu Nakano · 2026-05-29
arXiv:2605. 29371v1 Announce Type: cross Abstract: We study the subclass of potential mean-field games in which the running interaction cost and the terminal target cost are both expressed through reproducing-kernel maximum mean discrepancy (MMD) penalties, and develop a computational framework that exploits this kernel structure.
Read next because Kernel-based potential mean-field games with unbiased random Fourier $U$-statistics overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)". Matching terms: class, under, line, rate, control, trained. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29371v1 Announce Type: cross Abstract: We study the subclass of potential mean-field games in which the running interaction cost and the terminal target cost are both expressed through reproducing-kernel maximum mean discrepancy (MMD) penalties, and develop a computational framework that exploits this kernel structure. Both costs are estimated from finite-sample empirical distributions using a random Fourier U-statistic representation that is unbiased and has linear cost in the batch size. The drift of the controlled diffusion is parametrized by a neural network and trained via stochastic gradient descent. For this subclass we prove a sample-level almost-sure convergence theorem and an explicit almost-sure rate of convergence, under coupled rate conditions on the penalty parameter, the random-feature count, the sample size, and the optimization tolerance. The framework includes the kernel-MMD-penalty Schr\"odinger bridge problem as the special case of a vanishing interaction cost. Numerical experiments illustrate the method on the Schr\"odinger bridge problem in dimensions up to one hundred, and on an electric vehicle charging coordination problem with per-vehicle physical heterogeneity, where an aggregate-demand congestion cost represents price-feedback competition at the population level and the terminal MMD penalty shapes the state-of-charge distribution at the deadline.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses bias.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29272unread
Causal Label Recovery in Payment Networks
Gaurav Dhama · 2026-05-29
arXiv:2605. 29272v1 Announce Type: cross Abstract: Fraud detection models in payment networks train on chargeback labels that are systematically biased.
Read next because Causal Label Recovery in Payment Networks overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)". Matching terms: class, rect, correct, line, rate, stage, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29272v1 Announce Type: cross Abstract: Fraud detection models in payment networks train on chargeback labels that are systematically biased. Every label must survive three sequential gates: authorization (declined transactions generate no labels), issuer reporting (unreported fraud is invisible), and delay (pending chargebacks are missing at training time). Labels that do arrive may be corrupted by first-party misuse or issuer misclassification. A companion paper [arXiv:2605.27557] proved that these four impairments impose a minimax lower bound on detection performance. This paper asks: can that bound be achieved? We formalize the observation pipeline as a sequential missing-data problem with three propensity stages and a corruption layer, and construct the Sequential Triply Robust (STR) estimator. The STR corrects for all four impairments simultaneously and achieves the semiparametric efficiency bound -- no estimator can have lower asymptotic variance. It is sequentially triply robust: at each gate, consistency requires only that either the propensity model or the outcome regression is correctly specified, not both. We provide corruption correction via noise-rate-adjusted pseudo-labels, empirical Bayes shrinkage to stabilize inverse-propensity weights for small issuers, a plug-in variance estimator yielding valid confidence intervals, and a Bernstein concentration inequality for finite-sample guarantees. On the operational side, we derive the optimal training delay -- the maturity window that minimizes the sum of label-quality loss and model staleness -- and prove that the STR permits training on data that is days old rather than months old, decoupling model freshness from the chargeback maturity cycle. The STR provably dominates naive chargeback-based training in mean squared error for any sample size.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses bias.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29152unread
Do Deep Networks Forget Initialization? A Forgetting-Time View of Practical Inductive Bias
Mohua Das, Pierfrancesco Beneventano, Shibshankar Dey, Gareth H. McKinkey, Tomaso Poggio · 2026-05-29
arXiv:2605. 29152v1 Announce Type: cross Abstract: Randomly initialized neural networks induce a prior over functions, but the predictor used in practice is produced only after training.
Read next because Do Deep Networks Forget Initialization? A Forgetting-Time View of Practical Inductive Bias overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, epochs, line, rate, control, alone, trained, test. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29152v1 Announce Type: cross Abstract: Randomly initialized neural networks induce a prior over functions, but the predictor used in practice is produced only after training. We ask how much of this initial bias survives the training pipeline. To make the question measurable, we introduce initialization memory: the dependence of the validation-selected predictor on the scale of the random initialization. We perform controlled CIFAR-10 experiments on ResNets where initialization memory already sharply separates training regimes. Low-learning-rate SGD can interpolate while still remembering its initialization: on ResNet-9 with batch size $b=128$, test accuracy varies by $26.5$ percentage points across initialization scales despite $\ge99.5\%$ training accuracy. This is not undertraining: extending the same low-learning-rate regime to $5{,}000$ epochs leaves the spread essentially unchanged. In contrast, Adam-family methods largely erase the dependence. SGD can also be made to forget when larger learning rates are paired with explicit $L_2$ norm control. We interpret these findings in terms of the time scale of forgetting: gradient-flow-like dynamics can preserve initialization memory, whereas stochastic finite-step effects, explicit norm decay, and adaptive preconditioning erase it on scales governed by the size of explicit or implicit regularization. The practical inductive bias of a trained network is therefore not the architectural prior alone, but the architectural prior after being filtered by the forgetting dynamics of the training pipeline; and the same regularizers that improve generalization are precisely those that erase memory of initialization.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses bias.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29032unread
Theoretical Foundations and Effective Algorithms for Policy-Aware Simulator Learning
Christoph Dann, Yishay Mansour, Mehryar Mohri · 2026-05-29
arXiv:2605. 29032v1 Announce Type: cross Abstract: Model-based reinforcement learning (MBRL) agents typically learn world models by minimizing predictive loss.
Read next because Theoretical Foundations and Effective Algorithms for Policy-Aware Simulator Learning overlaps with clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone", experiment "Follow-up to #354: cascading chunk-binding — does A→B, B→C, C→D propagate the full chain on a recipient trained only to emit A?". Matching terms: line, rate, control, trained, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29032v1 Announce Type: cross Abstract: Model-based reinforcement learning (MBRL) agents typically learn world models by minimizing predictive loss. However, powerful RL optimizers inevitably exploit minor model inaccuracies, leading to simulator exploitation and a reality gap where policies succeed in simulation but fail in the real world. We propose that the objective for learning simulators should be strategic robustness rather than predictive accuracy, and formulate this as a zero-sum minimax game between a model player and an adversarial policy player. We provide a comprehensive theoretical analysis: (1) an online learning guarantee showing the game is learnable with sublinear regret bounds; (2) a tractable critic-based simplification bounding the global policy-value gap by the local critic's loss; and (3) an Error-MDP duality, proving that finding the worst-case policy is formally dual to a standard RL problem where the reward is the one-step critic error. This duality yields a provably convergent active data selection algorithm. Experiments on continuous control tasks demonstrate that our approach reduces prediction error in strategically important regions by $1.5$-$2.2\times$ and enables policies trained purely in simulation to match near-optimal real-world performance.
Potential threat/caveat for clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)": this item discusses robustness, adversarial.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.28894unread
Saddle Networks: Structure-Preserving Architectures for Convex-Concave Functions
Xavier Warin · 2026-05-29
arXiv:2605. 28894v1 Announce Type: cross Abstract: Saddle-point models arise throughout optimization, optimal transport, robust learning, and control.
Read next because Saddle Networks: Structure-Preserving Architectures for Convex-Concave Functions overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone", experiment "#351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation". Matching terms: under, control, position, test, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.28894v1 Announce Type: cross Abstract: Saddle-point models arise throughout optimization, optimal transport, robust learning, and control. In many applications, the relevant function f(x,y) is convex in x and concave in y, and preserving this geometry is essential for obtaining tractable min--max formulations and reliable certificates. We introduce a structured separable decomposition that preserves the convex-concave geometry and prove a complete one-dimensional approximation theorem under a mixed Monge-type convexity condition. We then describe practical saddle network architectures that preserve convexity in x and concavity in y by construction. The proposed architectures require only convexity-preserving neural networks, together with simple output transformations enforcing sign and concavity constraints. Finally, we report numerical benchmarks in dimension 1 and 5, showing that the proposed saddle networks achieve high accuracy on smooth, nonsmooth, and high-rank convex--concave test functions.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses benchmark.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29748unread
Instance-dependent Stochastic Lipschitz bandit
Marius Potfer, Vianney Perchet · 2026-05-29
arXiv:2605. 29748v1 Announce Type: new Abstract: We study the Lipschitz bandit problem, where a learner sequentially maximizes an unknown Lipschitz function $f$ over a domain $\mathcal{X} \subset [0,1]^d$ using noisy pointwise evaluations.
Read next because Instance-dependent Stochastic Lipschitz bandit overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: class, eval, rate, full. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29748v1 Announce Type: new Abstract: We study the Lipschitz bandit problem, where a learner sequentially maximizes an unknown Lipschitz function $f$ over a domain $\mathcal{X} \subset [0,1]^d$ using noisy pointwise evaluations. Existing regret bounds are either worst-case, scaling as $\tilde{\Theta} \left ( T^{d+1/d+2}\right )$, or adaptive via the zooming dimension $d_z$, yielding $\tilde{\Theta} \left ( T^{d_z+1/d_z+2}\right )$. However, such zooming-based guarantees are only partially instance-dependent, as they depend solely on the asymptotic growth of near-optimal level sets and fail to capture finer structural properties of $f$. We provide an analysis and an algorithm that characterizes the regret through integrals of the suboptimality gap of $f$ over its level sets. This yields regret bounds that adapt to the local growth of level sets, rather than only their asymptotic behavior. As a corollary, when the set of maximizers has dimension $d^\star>0$, we obtain improved adaptive rates of order $\tilde{\mathcal{O}} \left ( T^{d_z+1 / \max(d_z,d^\star)+2}\right )$ strictly improving over classical zooming bounds in this regime. Finally, we extend our analysis to the full-information setting (Lipschitz experts) and show how some of the regularity assumptions can be relaxed.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses evaluation.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29249unread
Prediction-Powered Inference Across Many Tasks for AI Evaluation & Social Science Research
Nicolas Emmenegger, Ellery Stahler, Chara Podimata · 2026-05-29
arXiv:2605. 29249v1 Announce Type: new Abstract: Many applications require statistically valid inference across many related tasks, while using only a handful of high-quality labels per hypothesis.
Read next because Prediction-Powered Inference Across Many Tasks for AI Evaluation & Social Science Research overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: rect, width, eval, line, rate, language, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29249v1 Announce Type: new Abstract: Many applications require statistically valid inference across many related tasks, while using only a handful of high-quality labels per hypothesis. In AI evaluation, these tasks may correspond to model behaviors across prompts, subgroups, or hypotheses; in social science surveys, they may correspond to related questions, populations, or measurement conditions. Prediction-powered inference (PPI) uses abundant but inexpensive proxy measurements to improve inference from limited, ground-truth labels, but commonly used methods treat tasks independently and therefore fail to exploit shared structure across related tasks. This limitation is especially important in settings where only a small number of labels are available per task. To address this issue, we introduce a multi-task prediction-powered inference framework that uses labeled data from related tasks to improve power while preserving task-specific inference. Our methods exploit the shared structure in the proxy-ground-truth relationship through cross-task recalibration, while retaining within-task rectification and power tuning to construct accurate point estimates and confidence intervals. We prove that efficiency gains beyond power-tuned PPI are only possible when the proxy-ground-truth relationship contains nonlinear structure; affine cross-task recalibrations are asymptotically equivalent to using the original proxy. We complement our theoretical findings with experiments on synthetic and semi-synthetic datasets, as well as a case study auditing language models on election-related information during the 2024 U.S. presidential election. Using a large human-annotation study, we show that cross-task recalibration can substantially reduce confidence interval widths when labels are scarce.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses limitation, evaluation.
source ↗open
score 100arxiv stat.ML (Machine Learning)arxiv:2605.29139unread
Anytime-Valid Federated Conformal RAG for LLM Swarms
Prasanjit Dubey, Xiaoming Huo · 2026-05-29
arXiv:2605. 29139v1 Announce Type: new Abstract: Federated Conformal RAG (FC-RAG) provides distribution-free coverage for a bandwidth-limited swarm of weak language models, but only at a fixed horizon.
Read next because Anytime-Valid Federated Conformal RAG for LLM Swarms overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: under, width, good, eval, fires, rate, control, position. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29139v1 Announce Type: new Abstract: Federated Conformal RAG (FC-RAG) provides distribution-free coverage for a bandwidth-limited swarm of weak language models, but only at a fixed horizon. We extend it to anytime-valid sequential coverage: validity at every stopping time, preserved under predictable adaptive control (recalibration, per-node bandwidth escalation, distilled-student refresh), at no extra cost in assumptions over fixed-horizon FC-RAG. Naive composition fails because FC-RAG's marginal coverage bound makes the betting e-process a non-supermartingale on adverse calibration draws, and Ville's inequality cannot be invoked. We give Anytime-FC-RAG, a sequential extension built on a summable per-step calibration-deviation budget that converts the marginal bound into a strict conditional bound on a calibration-good event, paired with a truncated betting e-process that is a nonnegative supermartingale on the entire probability space. From these two ingredients, we obtain four guarantees: time-uniform alarm validity $\mathbb{P}(\sup_t E_t \ge 1/\delta_e) \le \delta_e + \delta_{\mathrm{cal}}$, a Hoeffding-stitched cumulative-miscoverage envelope at the same total budget, safety under any predictable controller (recalibration, bandwidth escalation, student refresh), and training-side error propagation across an unbounded sequence of Federated Probe-Logit Distillation (FPLD) refreshes via a summable training budget. As a practical consequence, an adaptive controller that escalates retrieval bandwidth only when the e-process crosses a warning threshold matches the alarm rate of a fixed-high-bandwidth schedule at substantially lower communication cost. Experiments on a GPT-2-small + MiniLM swarm across MMLU, DBpedia, and AG News verify the predicted alarm rate, detection delay, envelope coverage, and $14$-$57\%$ bandwidth savings; the alarm fires when and only when coverage genuinely breaks.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses negative.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29651unread
Scarcity Is Not Enough: An Impossibility Result for Linear Sybil Cost Under Parallelizable Resources
Homayoun Maleki, Nekane Sainz, Jon Legarda, Igor Santos-Grueiro · 2026-05-29
arXiv:2605. 29651v1 Announce Type: new Abstract: Permissionless systems resist Sybil attacks by binding influence to scarce resources.
Read next because Scarcity Is Not Enough: An Impossibility Result for Linear Sybil Cost Under Parallelizable Resources overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: class, rect, under, source, line, rate, binding, alone. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29651v1 Announce Type: new Abstract: Permissionless systems resist Sybil attacks by binding influence to scarce resources. We show that scarcity alone is insufficient: the structural properties of the resource determine whether influence can be concentrated at sublinear cost through identity replication, delegation, or pooling. We model this through the adversarial cost C(s,T): the minimum expenditure required to achieve influence proportional to s independent participation units over T windows. We prove that any resource satisfying divisibility, additivity of influence, temporal reusability, and identity transferability admits influence amortization: C(s,T)=o(sT), regardless of protocol design. This is an impossibility result: no protocol rule can enforce linear cost of influence concentration over a structurally parallelizable resource. We further prove that throughput-bounded, non-transferable, window-local resources enforce C(s,T)=Omega(sT): each additional unit of sustained influence incurs marginal cost Delta(s,T)=Omega(T), growing with the time horizon. The two resource classes are asymptotically separated. As a direct design consequence, any mechanism targeting linear cost of influence concentration must ground participation in a resource that violates at least one parallelizability property.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses adversarial.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29620unread
Control Flow Graph Recovery for Dynamically Loaded Code via Symbolic Library Resolution
Oleksandr Mostovyi · 2026-05-29
arXiv:2605. 29620v1 Announce Type: new Abstract: Control Flow Graphs are one of the main data sources for software analysis that use dynamic and static software analysis methods.
Read next because Control Flow Graph Recovery for Dynamically Loaded Code via Symbolic Library Resolution overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, rect, soft, eval, source, compare, control, alone. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29620v1 Announce Type: new Abstract: Control Flow Graphs are one of the main data sources for software analysis that use dynamic and static software analysis methods. Protected software and modern malware increasingly depend on dynamic code loading techniques to evade static analysis. Usage of runtime dynamic linking mechanisms introduces unresolved indirect calls that stop static Control Flow Graph recovery. This serves to hide dynamic library that can be used for prevention of security analysis. To address this limitation, an analysis technique is proposed that combines symbolic execution with speculative library preloading to recover Control Flow Graphs from binaries by using dynamic loading. The methodology uses custom software hooks that intercept dynamic loading operations during symbolic execution and perform actual library loading into the analysis state. The module is based on a two-level architecture that stores interception functions and instruction tracking at the same time, all within a symbolic execution environment. To avoid executing potentially malicious code that dynamic instrumentation tools require, the analysis was conducted entirely through symbolic execution, making it safe for malware analysis. For evaluation a batch of 16 synthetic benchmarks was used, employing various obfuscation techniques including encrypted library names, network-triggered loading, environment-derived paths, multi-stage decryption chains, fileless execution and manual executable and linkable format parsing. The experiments results show that module recovers on average 29.8 % additional Control Flow Graph nodes and 26.5 % additional edges compared to static analysis alone, achieves 100 % precision and 100 % recall in library detection, with all discoveries validated through Frida-based dynamic instrumentation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, evaluation, benchmark.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29569unread
LoRA-Key: User-Centric LoRA Watermarking for Text-to-Image Diffusion Models
Yaopeng Wang, Qingliang Wang, Zhibo Wang, Huiyu Xu, Jiacheng Du, Qiu Wang, Jia-Li Yin, Kui Ren · 2026-05-29
arXiv:2605. 29569v1 Announce Type: new Abstract: Low-Rank Adaptation (LoRA) has become a widely used mechanism for customizing text-to-image diffusion models, enabling lightweight modules that are shared, reused, and commercialized as independent assets.
Read next because LoRA-Key: User-Centric LoRA Watermarking for Text-to-Image Diffusion Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, rect, under, line, project, without, alone, position. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29569v1 Announce Type: new Abstract: Low-Rank Adaptation (LoRA) has become a widely used mechanism for customizing text-to-image diffusion models, enabling lightweight modules that are shared, reused, and commercialized as independent assets. This LoRA-centric ecosystem shifts copyright protection from foundation models to distributed LoRA modules, which are easy to copy, redistribute, or reuse without authorization. Existing watermarking methods either protect the base diffusion model or require watermark-aware retraining for each target LoRA, limiting their practicality in open community settings. To address this limitation, we propose LoRA-Key, a user-centric LoRA watermarking framework that treats copyright protection as a reusable ownership key. LoRA-Key encapsulates a recoverable secret message into a standalone user-specific Watermark LoRA, which can be attached to different target LoRAs through training-free linear superposition without per-LoRA retraining or structural modification. To train such a reusable key, we first establish a latent watermark prior in the frozen VAE latent space for robust message embedding and recovery, and then optimize the Watermark LoRA with message-conditioned watermark supervision and semantic consistency constraints. We further introduce Gradient Orthogonal Projection (GOP) to suppress watermark updates that conflict with semantic-preserving directions, reducing interference with generation fidelity and downstream style adaptation. Extensive experiments show that LoRA-Key provides lightweight plug-and-play copyright protection while preserving generation quality and style fidelity, and maintains robust ownership verification under image-level distortions, downstream fine-tuning, and multi-LoRA composition.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29526unread
Temporal Motif-aware Graph Test-time Adaptation for OOD Blockchain Anomaly Detection
Runang He, Tongya Zheng, Huiling Peng, Yuanyu Wan, Bingde Hu, Jiawei Chen, Canghong Jin, Mingli Song, Can Wang · 2026-05-29
arXiv:2605. 29526v1 Announce Type: new Abstract: Ever-evolving transaction patterns have significantly hindered anomaly detection on emerging cryptocurrency blockchains due to the vast number of addresses and diverse anomalous behaviors.
Read next because Temporal Motif-aware Graph Test-time Adaptation for OOD Blockchain Anomaly Detection overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, text, rate, chain, test. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29526v1 Announce Type: new Abstract: Ever-evolving transaction patterns have significantly hindered anomaly detection on emerging cryptocurrency blockchains due to the vast number of addresses and diverse anomalous behaviors. Recently, advanced Graph Anomaly Detection (GAD) approaches applied to blockchains have faced two critical challenges: \textit{adversarial pattern evolution by malicious actors} and \textit{the out-of-distribution (OOD) problem caused by varied transaction semantics on blockchains}. To address these challenges, we propose a novel framework termed \textbf{TE}mporal \textbf{M}otif-aware \textbf{G}raph \textbf{T}est-\textbf{T}ime \textbf{A}daptation (\textbf{TEMG-TTA}). First, we comprehensively capture the 3-node temporal motif distribution of each active address using an efficient computational mechanism, enabling downstream temporal motif-aware graph learning. Second, we design a simple yet effective test-time adaptation strategy to facilitate the sharing of common patterns between training and testing graphs. Extensive experiments on 5 real-world datasets demonstrate that our proposed \textbf{TEMG-TTA} outperforms \textit{state-of-the-art} GAD approaches by an average of 54.88\%. A further case study on interpretable motif patterns reveals that \textbf{TEMG-TTA} explicitly characterizes the complex transaction patterns of anomalous addresses, thereby verifying the effectiveness of our technical designs. Our code will be made publicly available https://github.com/LuoXishuang0712/TEMG-TTA/.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses adversarial.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29468unread
SciIntBench: Measuring LLM Compliance with Research Integrity Norms Under Adversarial Framing
Almene De Meran Meguimtsop, Maria Leonor Pacheco, Daniel E. Acuna · 2026-05-29
arXiv:2605. 29468v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used to support scientific work, but it is unclear whether they uphold responsible conduct of research (RCR) norms or help undermine them.
Read next because SciIntBench: Measuring LLM Compliance with Research Integrity Norms Under Adversarial Framing overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, under, alignment, eval, language, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29468v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly used to support scientific work, but it is unclear whether they uphold responsible conduct of research (RCR) norms or help undermine them. We introduce SciIntBench, an adversarial benchmark of 810 prompts across ten RCR categories and three scientific domains. Each scenario appears as an Overt Adversarial, Covert Adversarial, and Benign version, allowing us to jointly measure framing-sensitive refusal of misconduct and helpfulness on legitimate requests. We evaluate 16 commercial and open-weight LLMs from six providers (2024--2026), producing 12,960 responses. We find that scientific integrity alignment is strongly framing-sensitive: models refuse explicit misconduct far more reliably than covert violations, especially failing when misconduct is presented as a pressure-driven shortcut. Refusals vary by RCR category, with weaker boundaries around transparency, plagiarism, and fabrication.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses adversarial, benchmark.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29450unread
Protecting On-Device AI Inference: A Systematic Review of Attacks and Defence Mechanisms
Zisis Tsiatsikas, Alexandros Fakis, Georgios Karopoulos, Vasileios Kouliaridis, Marios Anagnostopoulos · 2026-05-29
arXiv:2605. 29450v1 Announce Type: new Abstract: The need for secure and private Artificial Intelligence (AI) and Machine Learning (ML) on edge and mobile devices has increased the necessity of protecting the architecture of these systems from threats to both security and privacy.
Read next because Protecting On-Device AI Inference: A Systematic Review of Attacks and Defence Mechanisms overlaps with experiment "Implement Chen et al. persona-vector extraction recipe and compare to project's centroid-difference recipe", experiment "Follow-up to #354: cascading chunk-binding — does A→B, B→C, C→D propagate the full chain on a recipient trained only to emit A?", experiment "Test FR↔IT bystander-spill symmetry at multi-seed + 5 phrasings — pooled-rate vs per-phrasing asymmetry from #239 fact-check". Matching terms: extraction, trained, symmetry, asymmetry, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29450v1 Announce Type: new Abstract: The need for secure and private Artificial Intelligence (AI) and Machine Learning (ML) on edge and mobile devices has increased the necessity of protecting the architecture of these systems from threats to both security and privacy. With an ever-increasing number of pre-trained AI models being used on mobile platforms for client-side inference, there are rising concerns about the risks associated with the theft/extraction of AI models, adversarial attacks on AI models, and data breaches. As a result of this trend, a variety of defence mechanisms have been proposed to protect against these threats. These include Trusted Execution Environments (TEEs), homomorphic encryption, obfuscation, and differential privacy, among others. However, current surveys largely focus on edge intelligence, which includes distributed training, and thus overlook security and privacy issues that are specific to on-device AI inference. To the best of our knowledge, this paper presents the first comprehensive review of threats and corresponding defence mechanisms targeting on-device inference. Our results show that the attack and defence literature are unbalanced: approximately one quarter of the surveyed attack papers focus on Intellectual Property (IP) attacks, whereas half of the defence solutions tackle the same issue. More importantly, some attack categories have no defence paper associated to them, such as adversarial attacks that account for roughly one third of the attack literature. This asymmetry between known attacks and available mitigations highlights clear opportunities for future research on securing on-device AI inference.
Potential threat/caveat for experiment "Implement Chen et al. persona-vector extraction recipe and compare to project's centroid-difference recipe": this item discusses adversarial.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29434unread
AliMark: Enhancing Robustness of Sentence-Level Watermarking Against Text Paraphrasing
Yuexin Li, Wenjie Qu, Linyu Wu, Yulin Chen, Yufei He, Tri Cao, Bryan Hooi, Jiaheng Zhang · 2026-05-29
arXiv:2605. 29434v1 Announce Type: new Abstract: Existing sentence-level watermarking methods enhance robustness to paraphrasing by anchoring watermarks in sentence semantics.
Read next because AliMark: Enhancing Robustness of Sentence-Level Watermarking Against Text Paraphrasing overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, text, phrase, under, alignment, prefix, line, rate. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29434v1 Announce Type: new Abstract: Existing sentence-level watermarking methods enhance robustness to paraphrasing by anchoring watermarks in sentence semantics. However, their prefix-based designs remain vulnerable to structural perturbations, such as sentence splitting and merging, which commonly arise under strong paraphrasers like DIPPER and GPT-3.5. To mitigate this issue, we propose AliMark, a framework that reformulates sentence-level watermarking as a bit sequence encoding and alignment problem between a potentially watermarked text and a secret bit sequence. Notably, our approach adopts a two-stage detection strategy: we generate multiple restructured text variants and adaptively align their extracted bit sequences with the secret bit sequence to minimize alignment cost. This multi-candidate alignment design naturally improves robustness to sentence merges and splits. Extensive experiments demonstrate that AliMark substantially outperforms state-of-the-art baselines under diverse paraphrasing attacks.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses robustness.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29354unread
Harmless Yet Harmful: Neutral Prompting Attacks for Stealthy Hallucination Steering in Agent Skills
Chia-Yi Hsu, Chia-Mu Yu, Chun-Ying Huang, Jun Sakuma · 2026-05-29
arXiv:2605. 29354v1 Announce Type: new Abstract: LLM-powered coding agents increasingly participate in software development workflows by generating code, selecting dependencies, and producing package installation commands.
Read next because Harmless Yet Harmful: Neutral Prompting Attacks for Stealthy Hallucination Steering in Agent Skills overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, soft, eval, without, does, chain, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29354v1 Announce Type: new Abstract: LLM-powered coding agents increasingly participate in software development workflows by generating code, selecting dependencies, and producing package installation commands. This creates a new software supply chain risk: when an agent hallucinates a non-existent package, an attacker may register the hallucinated name and later compromise users who install it. Existing package hallucination attacks and defenses primarily focus on naturally occurring hallucinations, targeted dependency steering, or post-hoc package validation. In this paper, we introduce \emph{Neutral Prompting Attack} (NPA), a highly stealthy attack paradigm in which semantically benign instructions, such as encouraging imagination and exhaustiveness, increase package hallucination propensity without containing explicit malicious intent. Unlike targeted dependency steering, NPA does not specify an attacker-chosen package. Instead, it shifts the model's dependency generation behavior toward more speculative package names. We evaluate NPA across multiple coding-oriented LLMs and package hallucination benchmarks. Our results show that NPA increases both \emph{Hallucination ASR} and \emph{Pip Install ASR}, changes the distribution of hallucinated package names, and evades existing static-analysis, LLM-based, and agent-based Skill defenses. These findings reveal that harmless-looking prompts can covertly manipulate hallucination behavior and create downstream software supply chain risks.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29269unread
HunterAgent: Neuro-Symbolic Attack Trace Reconstruction under Anti-Forensics
Guangze Zhao, Yongzheng Zhang, Weilin Gai, Hongri Liu, Yuliang Wei, Bailing Wang · 2026-05-29
arXiv:2605. 29269v1 Announce Type: new Abstract: Modern alert-triage systems reduce SOC burden by filtering false positives, but flagging a high-risk alert is only the start of incident response.
Read next because HunterAgent: Neuro-Symbolic Attack Trace Reconstruction under Anti-Forensics overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: latin, under, source, line, rate, chain, trained, length. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29269v1 Announce Type: new Abstract: Modern alert-triage systems reduce SOC burden by filtering false positives, but flagging a high-risk alert is only the start of incident response. Threat hunting requires reconstructing causal attack chains across heterogeneous, partially corrupted logs. Against APTs using anti-forensics (parent-PID spoofing, log wiping, fileless execution), provenance graphs split into disjoint subgraphs and fail. Unconstrained LLM agents fabricate causal links violating OS physics, producing fluent but forensically inadmissible narratives. We propose HunterAgent, a neuro-symbolic framework that reframes trace reconstruction as cost-bounded heuristic graph search under partial observability. It uses an asymmetric Generator-Verifier pipeline: the LLM proposes semantic hypotheses within a typed ontology, while a verifier grounds each via identifier-level collisions on surviving orthogonal telemetry. To resolve severed traces, we score hops using a calibrated cost combining semantic divergence and OS temporal potential; schema violations are hard-pruned. A length-discounted epistemic budget prevents inferential drift and forces graceful halting. Under strict LOFO cross-validation on three public benchmarks and an in-house 40-trace dataset, HunterAgent achieves 86.1% mean F1, outperforming the top agentic baseline by 26.7 F1 and KAIROS by 17.1 F1, while cutting path-level hallucination from 61.5% to 6.4%. Under 70% log wiping, recall drops but precision stays >=84%, with 95.7% halting safely. All results hold under the realistic assumption that at least one orthogonal telemetry source survives.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29245unread
Implicit Identity Technologies for LLMs: Fingerprinting and Watermarking across Datasets, Models, and Generated Content
Bing Liu, Shunping Wang, Yufan Zhu, Xinyi Yu, Jing Huang, Linkang Du, Hongbin Pei, Wei Luo · 2026-05-29
arXiv:2605. 29245v1 Announce Type: new Abstract: This paper presents a survey and taxonomy of LLM fingerprinting and watermarking for identity, ownership verification, provenance, and generated-content attribution.
Read next because Implicit Identity Technologies for LLMs: Fingerprinting and Watermarking across Datasets, Models, and Generated Content overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: rect, under, eval, rate, stage, language, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29245v1 Announce Type: new Abstract: This paper presents a survey and taxonomy of LLM fingerprinting and watermarking for identity, ownership verification, provenance, and generated-content attribution. Large language models (LLMs) require substantial investments in data, computation, and expertise, and are increasingly deployed in high-stakes settings, making it critical to protect LLM-related assets and trace their origins. Existing work has rapidly expanded across dataset provenance, model ownership, and generated-content detection, but the field remains fragmented: fingerprinting and watermarking are often used inconsistently, and methods are typically studied within isolated asset-specific settings. To address this gap, we introduce implicit identity as a unifying abstraction for verifiable but not directly observable identity signals in LLM systems. We distinguish fingerprinting as non-intrusive identity derived from intrinsic characteristics, and watermarking as intrusive identity deliberately embedded into data, models, or generated content. We then propose a lifecycle-based taxonomy that organises techniques across datasets, models, and generated content, and further separates them by verification semantics: similarity-based attribution and keyed verification. Finally, we establish an evaluation framework centred on identifiability, robustness, and deployability, summarising representative metrics under realistic access and transformation regimes. By unifying terminology, lifecycle stages, and evaluation objectives, this survey provides a structured foundation for studying LLM identity technologies and for developing more reliable mechanisms for asset protection and provenance.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses robustness, evaluation.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29237unread
Evolving Skill-Structured Attack Memory Enhances LLM Jailbreaking
Junke Zhang, Jianwei Wang, Sishuo Chen, Yizhang He, Qingshuai Feng, Zhengyi Yang · 2026-05-29
arXiv:2605. 29237v1 Announce Type: new Abstract: Jailbreak attacks on large language models (LLMs) aim to induce LLMs to produce content that they are expected to refuse.
Read next because Evolving Skill-Structured Attack Memory Enhances LLM Jailbreaking overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, text, latin, eval, line, rate, lora, language. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29237v1 Announce Type: new Abstract: Jailbreak attacks on large language models (LLMs) aim to induce LLMs to produce content that they are expected to refuse. Automated black-box jailbreak generation is especially important for safety evaluation, where the attacker observes only model outputs and needs to automatically search for effective adversarial prompts. Existing black-box jailbreak methods either depend on sample-wise heuristic search or leverage attack experience through accumulating strategy pools or method libraries, lacking a systematic organization and management of attack experience. To mitigate these drawbacks, we propose MemoAttack, a memory-driven black-box jailbreak framework with comprehensive attack memory modeling, evolution, and selection. Specifically, MemoAttack comprises three key designs: (1) Skill-Structured Memory Modeling, which abstracts accumulated attack experience into reusable skill-structured attack memory whose units pair attack skills with templates, evidence, and lifecycle state; (2) Lifecycle-Driven Memory Evolution, which evolves the memory through evidence-based probation, promotion, retirement, reactivation, elimination, and storage cleanup; and (3) Explore-Exploit Balanced Memory Selection, which balances reliable memory reuse with uncertainty-driven exploration via contextual Thompson Sampling. Experiments on AdvBench demonstrate that MemoAttack achieves an average attack success rate of 98.00%, outperforming the strongest baseline by 16.67 percentage points, while reducing request count by 45.9%. Moreover, MemoAttack continuously improves as memory accumulates over more samples.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses adversarial, evaluation.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29178unread
The Best-Laid SCHEMEs: Coordinated Sabotage and Monitoring in Multi-Agent Systems
Nikolay Radev, Lennart Haas, Benjamin Arnav, Pablo Bernabeu-P\'erez · 2026-05-29
arXiv:2605. 29178v1 Announce Type: new Abstract: As agentic coding systems decompose work across multiple model instances, a critical safety question is whether those instances can coordinate to achieve a hidden malicious objective while remaining aligned with user intent.
Read next because The Best-Laid SCHEMEs: Coordinated Sabotage and Monitoring in Multi-Agent Systems overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, under, soft, eval, source, without, alone, full. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29178v1 Announce Type: new Abstract: As agentic coding systems decompose work across multiple model instances, a critical safety question is whether those instances can coordinate to achieve a hidden malicious objective while remaining aligned with user intent. We introduce SCHEME, a benchmark of 17 task instances across 7 settings and 8 real open-source libraries, each pairing a legitimate software-engineering task with a covert side task. Every setting is designed so that no proper subset of agents can succeed alone: agents must decompose a shared sabotage plan, relay partial requirements under different communication topologies, and execute mutually consistent edits, testing genuine multi-agent coordination rather than individual capability. Evaluating with GPT 5.1 Codex and Gemini 3.1 Pro, we find coordinated sabotage is already practical, with Gemini completing the covert objective while succeeding on the legitimate task in 84\% of samples and Codex in 46\%. A failure analysis shows that recovery, not failure incidence, drives the model gap: both models suffer similar numbers of communication failures per trajectory, but Gemini recovers from $\sim$73\% of occurrences against Codex's $\sim$56\%. However, the agents' code edits remain detectable: a trusted monitor flags coordinated sabotage at FPR$\le$2\% in 99\%/68\% (Gemini/Codex) of attempts from code edits alone, rising to 100\%/81\% with access to agent communication. Despite stealth-aware prompting, undetected sabotage stays at $<$3\% under the full-visibility monitor; telling agents a monitor is present halves Gemini's sabotage success without significantly breaking detection, even when agents are explicitly prompted to fabricate cover stories.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures, benchmark.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29177unread
EvaluatAR: A Cross-Device Evaluation Framework for Rapid Prototyping of Bystander PETs in AR
Syed Ibrahim Mustafa Shah Bukhari, Matthew Corbett, Bo Ji, Brendan David-John · 2026-05-29
arXiv:2605. 29177v1 Announce Type: new Abstract: Augmented Reality (AR) headsets continuously sense their surroundings, capturing nearby bystanders and raising privacy risks.
Read next because EvaluatAR: A Cross-Device Evaluation Framework for Rapid Prototyping of Bystander PETs in AR overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, eval, line, rate, control, stage, bystander. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29177v1 Announce Type: new Abstract: Augmented Reality (AR) headsets continuously sense their surroundings, capturing nearby bystanders and raising privacy risks. Visual bystander privacy-enhancing technologies (PETs) mitigate this risk by detecting bystanders in egocentric scene views and applying privacy transformations (e.g., obfuscation). However, traditional PET evaluation is human-dependent, high-overhead, and device-specific, making it difficult to reproduce across devices. We present EvaluatAR, a cross-device evaluation framework for rapid prototyping at the early stage of PET evaluation. Our framework enables controlled replication of experimental conditions by standardizing PET inputs (sensor data and visual stimuli) and outputs through a record-replay workflow. We validate EvaluatAR through three case studies on HoloLens 2, Magic Leap 2, and Meta Quest 3 across implicit (continuous, context-driven) and explicit (intent-driven) PETs: (1) cross-device replay of inputs to a PET to reveal device-specific privacy-performance trade-offs; (2) generalizability of the same framework workflow across implicit and explicit PET design categories; and (3) replay of privacy-relevant edge cases to diagnose failures and validate PET modifications, yielding an improvement over the state-of-the-art baseline. These results demonstrate EvaluatAR's support for rapid, iterative PET development to advance reproducible cross-device evaluation of bystander PETs at a critical moment in the emergence of ubiquitous AR.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures, evaluation.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29115unread
unix-ctf: Procedural Environments for Unix-Competence Reinforcement Learning
Geoffrey Bradway, Roger Creus Castanyer, Lorenz Wolf, Maxwill Lin, Matthew James Sargent, Augustine N. Mavor-Parker · 2026-05-29
arXiv:2605. 29115v1 Announce Type: new Abstract: Unix competence is the ability to use shell and operating-system primitives as first-class tools, not merely to write programs through a terminal.
Read next because unix-ctf: Procedural Environments for Unix-Competence Reinforcement Learning overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, text, class, rect, under, eval, token, line. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29115v1 Announce Type: new Abstract: Unix competence is the ability to use shell and operating-system primitives as first-class tools, not merely to write programs through a terminal. Current terminal benchmarks tend to blur this distinction: a solver fluent in Python but weak in Unix can pass a substantial fraction of Terminal-Bench 2.0, while the reverse skill profile is rarely exercised. We make the distinction operational and build a training surface for the Unix component. unix-ctf is a procedural generator of capture-the-flag tasks for shell agents. Each task hides a short token (a flag of the form flag(a3b1c9...)) inside a fresh Linux container using a single Unix feature, and the agent must recover it. Tasks are produced by an LLM-assisted synthesis pipeline that generates candidate hiding techniques, rewrites them into parameterized hide-and-find script pairs, and filters them with a bidirectional contract: the hide script must leave no plaintext trace of the flag on disk, and the find script must recover the flag in a fresh directory. Because the LLM only writes the planting and recovery steps (the container, layout, and grading harness are fixed), the pipeline lands 656 of 750 raw attempts as portable, reusable variants (87.5\%). Our reproduction of Endless Terminals' full-container-generation approach lands only 17.4\% under the same checks. The 656 variants canonicalize to 155 distinct techniques. Fine-tuning Qwen3-8B with LoRA using GRPO on this surface lifts solve rate from 11.6\% to 43.6\% on a 15-skill multi-family holdout (n=225), redistributes which InterCode-CTF tasks the model solves, and produces a +33 pp gain in Forensics while reaching 32/100 on InterCode-CTF. These results suggest that Unix competence is separable, trainable, and best evaluated directly rather than folded into programming-through-a-shell.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29114unread
ReasonBreak: Probing Vulnerabilities in Reasoning-Enabled Vision-Language-Action Models for Autonomous Driving
Mohammadreza Teymoorianfard, Jean-Philippe Monteuuis, Jonathan Petit, Amir Houmansadr · 2026-05-29
arXiv:2605. 29114v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models with integrated reasoning have been proposed for end-to-end autonomous driving, assuming a tight coupling between reasoning and trajectory generation.
Read next because ReasonBreak: Probing Vulnerabilities in Reasoning-Enabled Vision-Language-Action Models for Autonomous Driving overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, under, eval, rate, language, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29114v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models with integrated reasoning have been proposed for end-to-end autonomous driving, assuming a tight coupling between reasoning and trajectory generation. However, the robustness of such systems under realistic input perturbations remains largely unexplored. We show that these models are highly vulnerable to realistic input perturbations, achieving up to 89% attack success rate (ASR) on reasoning and up to 72% on trajectory manipulation in closed-loop simulation, leading to increased collision rates and degraded safety metrics. Using NVIDIA's recent Alpamayo models as representative industry-developed VLAs, we conduct the first systematic black-box study of reasoning-enabled VLA models under realistic textual input corruptions, evaluating their impact on reasoning and driving behavior. We introduce a reasoning-aware evaluation framework capturing both semantic and structural aspects of reasoning, along with safety-centric measures. We also introduce a benchmark for evaluating attacks and defenses on reasoning-trajectory interactions in autonomous driving. Our results highlight the need for rigorous evaluation and improved defenses to ensure the safety of reasoning-enabled VLA systems in autonomous driving.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses robustness, evaluation, benchmark.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.29107unread
GEO-Bench: Benchmarking Ranking Manipulation in Generative Engine Optimization
Ojas Nimase, Zhe Chen, Gengpei Qi, Yue Zhao, Xiyang Hu · 2026-05-29
arXiv:2605. 29107v1 Announce Type: new Abstract: Large language models (LLMs) increasingly rank products, documents, and recommendations for user queries, which makes manipulating these rankings a growing concern for fairness and information integrity.
Read next because GEO-Bench: Benchmarking Ranking Manipulation in Generative Engine Optimization overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, word, latin, rect, under, alpha, eval, rate. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.29107v1 Announce Type: new Abstract: Large language models (LLMs) increasingly rank products, documents, and recommendations for user queries, which makes manipulating these rankings a growing concern for fairness and information integrity. Research on generative engine optimization (GEO) has produced many manipulation methods, but each is evaluated on its own dataset with its own metrics, so their relative strength and detectability stay unclear. We present GEO-Bench, a benchmark that evaluates GEO ranking-manipulation attacks under one protocol. It unifies black-box prompt-based attacks (TAP, Zero-Shot), white-box gradient-based attacks (STS, RAF, StealthRank), and ten white-hat C-SEO strategies. We score every method on five datasets against a fixed open-weight ranker (Llama-3.1-8B-Instruct), using metrics for both effectiveness (NRG, Success@{\alpha}, Promote@{\alpha}) and stealth (keyword violation rate, perplexity ratio). Our evaluation shows that effectiveness and stealth trade off across adversarial attacks, that black-box content rewriting matches or exceeds gradient-based attacks on rank promotion while producing more fluent text and can evade both keyword- and perplexity-based detection on some domains, and that the access model does not predict attack strength. By standardizing datasets, attack implementations, and metrics, GEO-Bench enables the first direct comparison across these attack paradigms and supports the development of detection methods.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses adversarial, evaluation, benchmark.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.28914unread
AIRGuard: Guarding Agent Actions with Runtime Authority Control
Suliu Qin, Haomin Zhuang, Yujun Zhou, Yufei Han, Xiangliang Zhang · 2026-05-29
arXiv:2605. 28914v1 Announce Type: new Abstract: Tool-using language agents turn model decisions into external side effects: they read files, run scripts, call APIs, send messages, and invoke Model Context Protocol tools.
Read next because AIRGuard: Guarding Agent Actions with Runtime Authority Control overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, text, rect, source, compare, control, without, language. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.28914v1 Announce Type: new Abstract: Tool-using language agents turn model decisions into external side effects: they read files, run scripts, call APIs, send messages, and invoke Model Context Protocol tools. This makes agent attacks different from jailbreaks. The harmful step is often not an obviously forbidden output, but an ordinary executable action that becomes unsafe because attacker-controlled context steers authorized access against the user's interest. We identify this failure mode as authority confusion: untrusted resources may inform reasoning, but they must not authorize side effects. We present AIRGuard, a runtime guard that operationalizes least privilege as action-time authorization. AIRGuard normalizes heterogeneous tool calls, derives task authority into step-level authority, tracks source and target trust, simulates sensitive side effects, audits cross-step risk, and enforces decisions before actions execute. On AgentTrap, AIRGuard reduces Sonnet 4.6 attack success from 36.3% without defense to 5.5%. On DTAP-150, AIRGuard preserves 76.0% benign utility with Haiku 4.5, compared with 52.0% for ARGUS and 42.0% for MELON. An ablation further shows that prompt-only policy helps only modestly, whereas a dedicated runtime authority-control layer gives the agent system direct control over tool-mediated side effects. Code and data are available at https://github.com/Sophie508/AIRGuard.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.28899unread
Quantum-Enhanced Adversarial Robustness in Artificial Intelligence
Jaydip Sen · 2026-05-29
arXiv:2605. 28899v1 Announce Type: new Abstract: Artificial Intelligence has achieved remarkable success across diverse application domains.
Read next because Quantum-Enhanced Adversarial Robustness in Artificial Intelligence overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: class, rect, rate, full, position, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.28899v1 Announce Type: new Abstract: Artificial Intelligence has achieved remarkable success across diverse application domains. However, its vulnerability to adversarial attacks poses significant challenges to reliability, security, and trustworthiness. Adversarial machine learning demonstrates that even highly accurate models can be manipulated through carefully crafted perturbations, raising serious concerns in safety critical systems such as healthcare, finance, and autonomous technologies. In parallel, quantum computing has emerged as a transformative paradigm capable of addressing complex computational problems through principles such as superposition, entanglement, and quantum interference. The convergence of these fields has led to the emergence of quantum artificial intelligence, which explores how quantum techniques can enhance learning efficiency, scalability, and robustness. This chapter provides a comprehensive overview of adversarial machine learning and existing defense strategies, followed by an accessible introduction to quantum computing and quantum machine learning models. It further presents conceptual frameworks for quantum-enhanced adversarial robustness, emphasizing quantum optimization, feature mapping, and hybrid quantum classical architectures. Practical applications, key challenges, and future research directions are also discussed to support the development of secure and trustworthy AI systems.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses robustness, adversarial.
source ↗open
score 100arxiv cs.CR (Cryptography and Security)arxiv:2605.28890unread
Echoes within the Reasoning: Stealthy and Effective Watermarking via Chain of Thought
Jiacheng Lu, Yiming Li, Tao Song, Weijian Wang, Wenjie Qu, Haibing Guan, Jiaheng Zhang · 2026-05-29
arXiv:2605. 28890v1 Announce Type: new Abstract: Large Language Models with Chain-of-Thought reasoning capabilities represent valuable intellectual property, yet existing black-box watermarking methods often trade robustness for reasoning fidelity by perturbing final answers or relying on fragile trigger patterns.
Read next because Echoes within the Reasoning: Stealthy and Effective Watermarking via Chain of Thought overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, token, rate, control, without, chain, language, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2605.28890v1 Announce Type: new Abstract: Large Language Models with Chain-of-Thought reasoning capabilities represent valuable intellectual property, yet existing black-box watermarking methods often trade robustness for reasoning fidelity by perturbing final answers or relying on fragile trigger patterns. We propose BiCoT, a watermarking framework that embeds ownership signals into the internal geometry of reasoning traces by aligning high-saliency structural anchors with a private signature subspace while regularizing ordinary control tokens to preserve semantic capacity. This design couples the watermark with reasoning-relevant representations, making removal difficult without disrupting the features that support coherent reasoning. To enable verification under model theft and representation drift, we introduce Robust Subspace Registration (RSR), a Top- logprob-based black-box verifier that uses sentinel tokens to calibrate systematic shifts in the output distribution. Experiments show that BiCoT preserves reasoning fidelity across diverse complex reasoning tasks while achieving robust detection under fine-tuning, quantization, model-level perturbations, and adaptive output-level attacks across in-domain and out-of-distribution settings.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses robustness.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29225unread
BenchTrace: A Benchmark for Testing Reflection Ability and Controlled Evolution in LLM Agents
Jiahao Huang, Fei Cheng, Junfeng Jiang, Zefan Yu, Akiko Aizawa · 2026-05-29
arXiv:2605. 29225v1 Announce Type: new Abstract: Self-evolving agents improve over time by reflecting on past failures, but existing evaluation is limited in two ways: it measures only task scores, leaving reflection quality unknown, and it relies on agents' own episode runs, offering no mechanism to target specific failure patterns.
Read next because BenchTrace: A Benchmark for Testing Reflection Ability and Controlled Evolution in LLM Agents overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, text, rect, correct, eval, line, rate, control. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29225v1 Announce Type: new Abstract: Self-evolving agents improve over time by reflecting on past failures, but existing evaluation is limited in two ways: it measures only task scores, leaving reflection quality unknown, and it relies on agents' own episode runs, offering no mechanism to target specific failure patterns. We present \textbf{BenchTrace}, a benchmark for evaluating self-evolution ability in LLM agents. BenchTrace is built on a snapshot-reflection dataset of 1,821 annotated episodes spanning six diverse tasks, and comprises a \textbf{Reflection Evaluation} that probes failure identification through targeted QA tasks, and an \textbf{Evolution Evaluation} that tests whether past failure experience translates into avoidance behavior in a controlled self-evolution simulation. Building on BenchTrace, we propose \textbf{failure avoidance rate (FAR)}, a new evaluation metric measuring the fraction of test cases in which the agent successfully avoids the target failure instance. Experiments with Qwen3-32B and GPT-4.1 reveal that both models fall below a 30\% end-to-end pass rate on reflection evaluation, with diagnosis as the primary bottleneck. Evolution evaluation shows that self-evolution methods generally improve FAR over the non-evolving baseline, but agents forget early lessons as noise episodes accumulate, and agents fail to generalize their reflections beyond the specific context, causing negative transfer across task contexts. Our correlation analysis further reveals that only a fully correct reflection is strongly associated with higher FAR. BenchTrace exposes concrete limits of current self-evolution approaches and provides a controlled, model-agnostic framework for targeted evaluation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures, negative, evaluation, benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29218unread
GTA: Generating Long-Horizon Tasks for Web Agents at Scale
Tenghao Huang, Kung-Hsiang Huang, Prafulla Kumar Choubey, Yilun Zhou, Muhao Chen, Jonathan May, Chien-Sheng Wu · 2026-05-29
arXiv:2605. 29218v1 Announce Type: new Abstract: Web agents, which couple language models with browsing and tool-use capabilities, show promise as open web assistants.
Read next because GTA: Generating Long-Horizon Tasks for Web Agents at Scale overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, eval, assistant, line, rate, control, without, position. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29218v1 Announce Type: new Abstract: Web agents, which couple language models with browsing and tool-use capabilities, show promise as open web assistants. Yet progress is increasingly limited by the lack of scalable, process-level supervision. Existing benchmarks are largely manually constructed, providing only coarse start-goal annotations without intermediate trajectories, while recent automatic generation efforts remain expensive, biased, and shallow. These limitations prevent reliable training and evaluation of agents that must generalize to realistic, multi-hop, cross-page tasks. We introduce a scalable framework, GTA, that integrates crawling, retrieval-based seeding, in-context generation, and automated quality control to produce realistic tasks paired with executable trajectories. This design decouples crawling from generation for greater efficiency, grounds tasks in the site graph to enforce compositionality, and ensures dense supervision through deterministic replays and systematic validation. We instantiate the pipeline on over 50 websites covering e-commerce, government, forums, and news, with multilingual and multi-hop coverage. The resulting benchmark reveals a significant human-agent performance gap and enables detailed diagnostics. Our contributions are three-fold: (i) formalizing multi-hop web-agent task generation, (ii) proposing an efficient and validated pipeline for automatic data creation, and (iii) releasing a dynamic benchmark with reproducible evaluation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations, bias, evaluation, benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29192unread
ReasonOps: Operator Segmentation for LLM Reasoning Traces
Daniel Lee, Owen Queen, James Zou · 2026-05-29
arXiv:2605. 29192v1 Announce Type: new Abstract: Chain-of-thought traces from large reasoning models can span tens of thousands of tokens, yet we lack a vocabulary for describing their internal structure.
Read next because ReasonOps: Operator Segmentation for LLM Reasoning Traces overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, class, rect, correct, source, token, line, alone. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29192v1 Announce Type: new Abstract: Chain-of-thought traces from large reasoning models can span tens of thousands of tokens, yet we lack a vocabulary for describing their internal structure. Previous methods developed to analyze chain-of-thought traces are either too rigid or not expressive enough, failing to capture features across domains and models. To remedy this, we develop ReasonOps, an unsupervised, expressive method for annotating chain-of-thought traces, providing succinct universal operators. Using ReasonOps, we analyze 44,662 traces from 12 thinking LLMs spanning 6 families across 8 reasoning benchmarks and discover that they share a common compositional structure: 7 recurring reasoning operators -- discourse-level moves such as backtracking, inferring, and hypothesizing -- that emerge from unsupervised clustering of sentence-initial 3-token pivots. These operators appear across every model family and benchmark domain, confirmed by three independent LLM judges who classify held-out samples at 70 -76% accuracy. We analyze the structure of operators on easy vs. hard problems, revealing that reflective operators are more helpful on hard problems and harm performance on easy problems. Operator sequences are highly model-identifying: a classifier trained on operator distributions alone recovers the source model with macro-AUC, revealing that each model family has a distinctive reasoning fingerprint. Structural operator features predict within-problem answer correctness well above baselines. Classifiers built on these operators reach WP-AUC and on AIME specifically. ReasonOps further enables early quality estimation well before the trace completes: we predict at WP-AUC for only 50% of the trace. The ReasonOps pipeline is unsupervised and annotation-free, enabling deep insights into LLM reasoning traces as well as strong downstream results on model identification and correctness prediction.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29174unread
Paper Agents, Paper Gains: An Empirical Analysis of DeFi Investment Agents
Jay Yu, Amy Zhao, Danning Sui · 2026-05-29
arXiv:2605. 29174v1 Announce Type: new Abstract: DeFi investment agents, systems that use AI for autonomous on-chain trading, have attained over USD 3 billion in combined token valuations since late 2024.
Read next because Paper Agents, Paper Gains: An Empirical Analysis of DeFi Investment Agents overlaps with clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "A pretraining-data-poisoned Qwen3-4B backdoor only fires on the exact trigger tokens — paraphrases don't activate it, and base-model similarity to the trigger doesn't predict which inputs fire (MODERATE confidence)". Matching terms: alignment, token, rate, project, chain. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29174v1 Announce Type: new Abstract: DeFi investment agents, systems that use AI for autonomous on-chain trading, have attained over USD 3 billion in combined token valuations since late 2024. We survey over 1,900 AI-tagged crypto projects, filter to investment-focused agents, and curate 10 representative projects spanning strategy and observability dimensions. We then conduct a deep-dive architectural analysis of two prominent agent frameworks, ElizaOS and Virtuals Protocol, and a quantitative on-chain performance analysis of 11 Solana-based agent treasuries with publicly attributable trading activity, covering 925,323 token holders. We find that current deployments remain early and heterogeneous: (1) in our sample, many projects do not yet provide clear evidence of autonomous trade execution, and developer interviews suggest that many visible deployments remain basic API integrations; (2) agent treasuries retain over USD 30M in paper gains while token holders collectively lost USD 191.7M, with the top 1% of wallets capturing 81.4% of all gains (USD 1.81B); (3) token valuations are weakly connected to treasury fundamentals, with market-cap-to-AUM ratios exceeding 10,000x versus below 1x for established DeFi protocols; and (4) aggregate user gains peaked at USD 2.4B before declining to net losses, with median returns negative on every platform and tokens declining 93% on average from all-time highs. We interpret these outcomes as characteristic of a permissionless, first-generation market in which open infrastructure enables rapid experimentation but also allows naive or speculative agents to launch before robust standards for autonomy, performance, and stakeholder alignment emerge. We therefore propose a maturity framework along three dimensions: autonomous execution, risk-adjusted profitability, and stakeholder alignment, to characterize the gap between current deployments and future investment-grade agent systems.
Potential threat/caveat for clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)": this item discusses negative.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29123unread
The Confidence Shortcut: A Reasoning Failure Mode of Masked Diffusion Models
Dueun Kim, Albert No · 2026-05-29
arXiv:2605. 29123v1 Announce Type: new Abstract: Masked diffusion language models (MDMs) uniquely support any-order generation, with confidence-based decoding currently serving as the de facto standard inference policy.
Read next because The Confidence Shortcut: A Reasoning Failure Mode of Masked Diffusion Models overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: rect, alignment, rate, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29123v1 Announce Type: new Abstract: Masked diffusion language models (MDMs) uniquely support any-order generation, with confidence-based decoding currently serving as the de facto standard inference policy. To optimize for this, recent training schemes attempt to align training mask patterns directly with those observed during generation. However, we argue that confidence-based decoding is inherently misaligned with the logical-flow trajectories required for complex reasoning, and that confidence-aligned training actively entrenches this misalignment. We make this concrete using multi-digit addition, where the decoding strategy prematurely predicts locally easy digits before resolving their long-range dependencies, producing high-confidence errors on challenging inputs. While traditional random masking keeps the failure rate low on this challenging tail, confidence-aligned training amplifies the error rate by an order of magnitude. Across five distinct reasoning tasks, this same pattern emerges with task-dependent severity: confidence-based decoding induces failures on highly complex inputs, and confidence-aligned training exacerbates them. In contrast, random masking -- despite its perceived inefficiency -- robustly preserves the reasoning-trajectory conditionals essential for solving the challenging tail.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses failure, failures.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29119unread
PRO-CUA: Process-Reward Optimization for Computer Use Agents
Yifei He, Rui Yang, Hao Bai, Tong Zhang, Han Zhao · 2026-05-29
arXiv:2605. 29119v1 Announce Type: new Abstract: Computer use agents (CUAs) have shown strong potential for automating complex digital workflows, yet their training remains constrained by costly live environment interaction and limited high-quality supervision.
Read next because PRO-CUA: Process-Reward Optimization for Computer Use Agents overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, line, rate, without, trained, on-policy, candidate, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29119v1 Announce Type: new Abstract: Computer use agents (CUAs) have shown strong potential for automating complex digital workflows, yet their training remains constrained by costly live environment interaction and limited high-quality supervision. Existing filtered behavior cloning pipelines suffer from imitation bottlenecks, including distribution shift from the expert demonstration and the absence of negative learning signals. Meanwhile, standard trajectory-level reinforcement learning struggles with sparse rewards, ambiguous credit assignment, and high infrastructure costs for long-horizon GUI interaction. In this work, we propose PRO-CUA, a process-reward optimization framework for training CUAs with iterative step-level reinforcement learning. PRO-CUA decouples on-policy environment interaction from policy optimization: the current policy collects states through live rollouts, generates diverse candidate actions for each state, receives step-level feedback from a process reward model (PRM), and is optimized with group-relative advantages. This design enables dense and flexible credit assignment without relying on golden answers or offline expert trajectories, while reducing distribution shift by training on the agent's own execution states. Experiments on live web benchmarks demonstrate the effectiveness of PRO-CUA and the reliability of PRM-guided step-level training.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses negative, benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29087unread
The Chain Holds, the Answer Folds: Trace-Answer Dissociation in Reasoning Models Under Adversarial Pressure
Yubo Li, Ramayya Krishnan, Rema Padman · 2026-05-29
arXiv:2605. 29087v1 Announce Type: new Abstract: Reasoning models are evaluated on single-turn benchmarks but deployed in multi-turn dialogue, where users push back on correct answers.
Read next because The Chain Holds, the Answer Folds: Trace-Answer Dissociation in Reasoning Models Under Adversarial Pressure overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: rect, under, correct, wrong, eval, token, fires, line. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29087v1 Announce Type: new Abstract: Reasoning models are evaluated on single-turn benchmarks but deployed in multi-turn dialogue, where users push back on correct answers. Under sustained adversarial pressure we find a previously undocumented failure mode: the chain-of-thought stays factually correct from first turn to last while the emitted answer flips wrong. We call this unfaithful capitulation (UC) and isolate it with a $2\times 2$ latent-versus-behavioral framework that flip-rate metrics and single-turn faithfulness probes both miss. Across three datasets (MT-Consistency, MMLU-Pro, GSM8K), the latent-correct rate at the behavioral flip clusters near 50% in think mode and collapses to 11-15% under no_think -- paired, within-model causal evidence that reasoning creates the gap. Across models the effect tracks the reasoning channel (high in Qwen3-32B and GPT-OSS-20B, low in inline-CoT Gemma-4-31B-it). An independent GPT-4o judge corroborates $86\%$ of UC labels; a token-level probe shows the answer-slot argmax is correct in $84\%$ of UC cells; and a naive trace-anchored defense backfires. We release all trajectories, traces, and judge labels.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses failure, adversarial, benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29082unread
The Importance of Out-of-Band Metadata for Safe Autonomous Agents: The Redpanda Agentic Data Plane
Tyler Akidau, Tyler Rockwood, Johannes Br\"uderl, Marc Millstone · 2026-05-29
arXiv:2605. 29082v1 Announce Type: new Abstract: AI agents are increasingly expected to operate as digital employees: accessing enterprise data, making decisions, and taking actions autonomously.
Read next because The Importance of Out-of-Band Metadata for Safe Autonomous Agents: The Redpanda Agentic Data Plane overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, class, rate, cascading, propagate, full, stage. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29082v1 Announce Type: new Abstract: AI agents are increasingly expected to operate as digital employees: accessing enterprise data, making decisions, and taking actions autonomously. But agents are simultaneously less predictable than humans -- prone to hallucination, misinterpretation, and adversarial manipulation -- and more technically capable: with deep system knowledge and high-throughput interfaces cascading damage at machine speed. This combination makes it unsafe to rely on agents to faithfully interpret or propagate security-critical metadata such as access policies, data classifications, and behavioral constraints. We present the Redpanda Agentic Data Plane (ADP), an architecture built around out-of-band metadata channels: infrastructure pathways that carry security context, policy signals, and audit trails deterministically, entirely outside the agent's read and write path and across heterogeneous infrastructure. These channels enforce governance at every stage of the agent lifecycle -- scoping data access on the way in, constraining actions during execution, and capturing tamper-proof transcripts on the way out. We demonstrate ADP with a multi-agent portfolio rebalancing system in which autonomous agents monitor markets, make trade decisions, and execute orders across isolated client accounts -- with per-client data scoping, trade approval thresholds, and tamper-proof audit trails all enforced by out-of-band channels the agents can neither see nor bypass.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses adversarial.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29078unread
Bridging the Sim-to-Real Gap in Reinforcement Learning-Based Industrial Dispatching through Execution Semantics
Jonathan Hoss, Noah Klarmann · 2026-05-29
arXiv:2605. 29078v1 Announce Type: new Abstract: Event-driven scheduling policies are increasingly deployed in industrial environments, where decisions are made under asynchronous and partially observed system states.
Read next because Bridging the Sim-to-Real Gap in Reinforcement Learning-Based Industrial Dispatching through Execution Semantics overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, under, eval, full. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29078v1 Announce Type: new Abstract: Event-driven scheduling policies are increasingly deployed in industrial environments, where decisions are made under asynchronous and partially observed system states. As a result, decision states are not temporally consistent, action admissibility is not explicitly defined, and the origin of execution errors remains ambiguous. These issues limit both reliability and interpretability. To address this gap, a policy-neutral execution and measurement layer is proposed to mediate between scheduling policies and the industrial execution environment. The layer constructs decision-valid snapshots from asynchronous event streams, defines a standardized execution contract with explicit action admissibility, and records outcomes as divergences between policy intent, transactional outcomes, physical execution, and human intervention. This enables a separation between decision semantics and execution behavior and makes deployment mismatch observable and structurally attributable. The proposed framework is evaluated using a discrete-event simulation. The results show analytical benefits across all observation lag regimes, as undifferentiated execution failures are transformed into structured, typed outcomes with full attribution coverage. Operational benefits are strongest under low observation lag, where avoidable execution errors can be prevented before commitment. Overall, the layer turns execution uncertainty into supervisory data for evaluation and policy refinement.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures, evaluation.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29068unread
Robust and Efficient Guardrails with Latent Reasoning
Siddharth Sai, Xiaofei Wen, Muhao Chen · 2026-05-29
arXiv:2605. 29068v1 Announce Type: new Abstract: Maintaining the safety of large language models (LLMs) is crucial as they are increasingly deployed in real-world applications.
Read next because Robust and Efficient Guardrails with Latent Reasoning overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: class, rect, eval, token, line, stage, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29068v1 Announce Type: new Abstract: Maintaining the safety of large language models (LLMs) is crucial as they are increasingly deployed in real-world applications. Existing safety guardrails typically rely on single-pass classification or, more recently, distilled reasoning. Reasoning-based guardrails significantly outperform classification-only baselines, but they incur substantial query latency and token overhead that make them impractical for highthroughput deployment. To address this challenge, we propose COLAGUARD, a guardrail model that transfers multi-step safety reasoning into a continuous latent space through a stage-wise training curriculum, enabling direct hidden-state propagation at inference. Evaluated on ten prompt- and response-moderation settings spanning eight safety benchmarks, COLAGUARD improves macro-F1 by 8.24 points over Llama Guard 3 and matches our explicit reasoning baseline, GuardReasoner, in macroF1 while delivering a 12.9X speedup and 22.4X reduction in token usage. Our results suggest that latent reasoning offers a practical alternative to explicit rationale generation for deployable guardrails, jointly improving safety robustness and inference efficiency rather than treating them as competing objectives.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses robustness, benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29055unread
Hallucination Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Caching
Diego Gosmar, Deborah A. Dahl · 2026-05-29
arXiv:2605. 29055v1 Announce Type: new Abstract: Hallucination remains a major reliability barrier for production LLM systems, particularly in multi-agent pipelines where unsupported claims can propagate unchecked across stages.
Read next because Hallucination Mitigation with Agentic AI, Nested Learning, and AI Sustainability via Semantic Caching overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, text, rect, correct, eval, line, rate, without. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29055v1 Announce Type: new Abstract: Hallucination remains a major reliability barrier for production LLM systems, particularly in multi-agent pipelines where unsupported claims can propagate unchecked across stages. This paper adapts a HOPE-inspired Nested Learning architecture with Continuum Memory Systems (CMS) and semantic similarity caching to a hybrid benchmark of 310 prompts combining 217 epistemic-uncertainty prompts and 93 fabrication-induction stress-test prompts. A three-stage agentic pipeline orchestrated via the Open Floor Protocol (OFP) is evaluated with five KPIs -- FCD (Factual Claim Density), FGR (Factual Grounding References), FDF (Fictional Disclaimer Frequency), ECS (Explicit Contextualization Score), and OSR (Observability Score Ratio) -- aggregated into THS (Total Hallucination Score) across five weighting configurations to study mitigation-observability trade-offs. FDF, ECS, OSR, and FGR are subtracted as mitigation signals, so that a more negative THS indicates stronger mitigation. The FrontEndAgent is configured as a high-stochasticity generator (temperature = 1.0) to produce a realistic hallucination baseline, while the SecondLevelReviewer and ThirdLevelReviewer operate as progressive correctors. This asymmetric design yields end-to-end THS reductions of -31.3% to -35.9% across five weighting configurations. Semantic caching achieves 440 cache hits over 930 potential calls (47.3% hit rate), reducing LLM invocations to 490, lowering energy and CO2e footprint, and making multi-stage review pipelines operationally viable at production scale. ExtremeObservability attains the most negative final THS (-0.0709), confirming that observability-heavy configurations reinforce rather than compromise mitigation. These findings suggest that memory-augmented multi-agent designs can jointly improve factual reliability, operational efficiency, and auditability without model retraining.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses negative, benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29041unread
Practitioner Beliefs and Behaviors in AI-Enhanced Education: DOT Framework Survey Evidence
David Gibson (Curtin University), M. Elizabeth Azukas (Georgia Institute of Technology), Gerald Knezek (University of North Texas) · 2026-05-29
arXiv:2605. 29041v1 Announce Type: new Abstract: This study reports findings from a cross-sectional survey (n = 72) of higher education practitioners examining beliefs, behaviors, and institutional conditions related to artificial intelligence (AI) integration in teaching and learning.
Read next because Practitioner Beliefs and Behaviors in AI-Enhanced Education: DOT Framework Survey Evidence overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rect, alpha, eval, rate, implement, factor, lora. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29041v1 Announce Type: new Abstract: This study reports findings from a cross-sectional survey (n = 72) of higher education practitioners examining beliefs, behaviors, and institutional conditions related to artificial intelligence (AI) integration in teaching and learning. Grounded in the DOT Framework, which integrates design thinking and open systems theory, the study investigates AI familiarity, usage patterns, design-oriented practices, and pedagogical beliefs. Exploratory factor analysis of 19 belief items identified a three-factor structure: AI Functional Capabilities, Oversight and Governance, and Instructor Collaboration and Planning ({\alpha} = .90). Results indicate that practitioners hold favorable views of AI as a pedagogical support while maintaining strong commitments to human oversight and critical evaluation. Reported practices emphasize iterative prompting and content generation, with less consistent use of needs assessment and feedback loops. Institutional barriers including limited policy, training, and infrastructure were widely reported. These findings provide preliminary empirical support for the DOT Framework as a descriptive model of practitioner beliefs and practices, while also highlighting gaps between design-oriented theory and current implementation. The study contributes an initial measurement structure and identifies directions for confirmatory validation and outcome-based research linking AI-supported design practices to instructional quality.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29025unread
When Models Disagree: Rethinking LLM Evaluation for Public Comment Analysis
Aisha Najera, Alvin Moon, Vedant Srinivasan, Rajesh Veeraraghavan · 2026-05-29
arXiv:2605. 29025v1 Announce Type: new Abstract: Federal agencies are deploying large language models (LLMs) to categorize public comment corpora, where the model's organization of the record shapes what policymakers see and which arguments register.
Read next because When Models Disagree: Rethinking LLM Evaluation for Public Comment Analysis overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: rect, eval, line, without, stage, language, model, absent. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29025v1 Announce Type: new Abstract: Federal agencies are deploying large language models (LLMs) to categorize public comment corpora, where the model's organization of the record shapes what policymakers see and which arguments register. Standard evaluation, anchored on stance accuracy against a small validated set, cannot detect when different models produce materially different categorizations of the same public input. We propose an Interpretive Audit Pipeline that treats multi-model disagreement as diagnostic of interpretive complexity and directs human review toward genuinely ambiguous public input. Analyzing 1,260 public comments on a federal USDA docket across four LLMs, we find that inter-model thematic divergence exceeds within-model prompt variation, and that an expert rubric suppresses deep interpretive disagreement without resolving it. In a two-stage labeling study on a stratified 40-comment subsample, four LLMs and a human annotator labeled independently and then revised after seeing the others' labels. Revision behavior varied across labelers, and the human annotator's revisions frequently introduced framings absent from the ensemble's collective output. We argue disagreement-based evaluation is a necessary complement to accuracy metrics for LLM-assisted interpretive coding.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses evaluation.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.29018unread
Adopt $\neq$ Adapt: Longitudinal Analyses of LLM Conversations in the Wild
Rebecca M. M. Hicke, Kiran Tomlinson · 2026-05-29
arXiv:2605. 29018v1 Announce Type: new Abstract: Although a growing body of research has begun to describe user--LLM interactions, the picture it paints is largely static; little is known about how individual users change their behavior over time.
Read next because Adopt $\neq$ Adapt: Longitudinal Analyses of LLM Conversations in the Wild overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "Implement Chen et al. persona-vector extraction recipe and compare to project's centroid-difference recipe". Matching terms: soft, rate, compare, does. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.29018v1 Announce Type: new Abstract: Although a growing body of research has begun to describe user--LLM interactions, the picture it paints is largely static; little is known about how individual users change their behavior over time. To address this gap, we analyze the conversational trajectories of $\sim$12,000 randomly sampled Microsoft Bing Copilot users and compare these with data from WildChat-4.8M. While the Copilot data contains significant population-level trends, we find that trends in individual user trajectories are much weaker; user habits prove to be overwhelmingly sticky. We also find stark differences between users of different activity levels: more active users have more successful conversations and use the LLM for more complex and professionally oriented tasks. Some user trends also appear in WildChat-4.8M, but we find evidence that this dataset is significantly skewed towards highly proficient "power" users. Ultimately, our results suggest that existing user behavior is difficult to change and demonstrate the extent of user heterogeneity. Our comparison between datasets highlights that WildChat does not represent typical user-AI interactions, an important caveat for downstream uses of the data.
Potential threat/caveat for clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)": this item discusses caveat.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.28994unread
BEAMS: Benchmarking and Evaluating AI for Modeling and Simulation
Sara Metcalf, William Schoenberg · 2026-05-29
arXiv:2605. 28994v1 Announce Type: new Abstract: AI tools to support real world decision making must be able to build simulation models that inform their recommendations and render them interpretable.
Read next because BEAMS: Benchmarking and Evaluating AI for Modeling and Simulation overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: eval, source, rate, implement, project, test, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.28994v1 Announce Type: new Abstract: AI tools to support real world decision making must be able to build simulation models that inform their recommendations and render them interpretable. Tools that can automate aspects of modeling practice must complement human expertise, not replace it. The BEAMS Initiative aims to guide the development of AI tools for modeling and simulation toward forms that are responsible and ethical by establishing benchmarks for human centered modeling and simulation practices. The initiative uses open digital and organizational infrastructure to collaboratively evaluate AI tools for modeling and simulation. The open source sd ai project hosted by the initiative establishes transparency and enables contributions to be shared broadly. A steering group focuses on prioritizing potential benchmarks, while a technical group focuses on implementing the benchmarks in the form of automated tests. Tests for several distinct categories of evaluation have been implemented and applied to AI tools that support qualitative model building, quantitative model building, and model discussion. These include tests for causal translation, model iteration, causal reasoning, conformance, model behavior explanation, suggested model building steps, and suggested model fixes. When engines from the sd ai project are coupled with different LLMs, their performance on these evaluations reveals variability across different AI tools. The evaluations implemented by the initiative demonstrate that AI enabled modeling tools perform better at discussion and basic qualitative tasks than with causal reasoning and quantitative error fixing. No single LLM dominates across engine types, highlighting the importance of specific tasks and tradeoffs between speed and accuracy. Ongoing efforts of the initiative aim to incorporate benchmarks that address concerns about bias by considering alternative perspectives and human centered use cases.
Potential threat/caveat for clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)": this item discusses bias, evaluation, benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.28978unread
VFEAgent: A Multimodal Agent Framework for End-to-End Automated Finite Element Analysis
Jiachen Zhang (Peking University, China Agricultural University), Junyi Lao (Peking University), Chenghao Liu (Peking University), Siyuan Liu (Peking University), Shixin Wu (Peking University), Linsen Zhang (Peking University), Boyu Wang (Peking University), Songfang Huang (Peking University) · 2026-05-29
arXiv:2605. 28978v1 Announce Type: new Abstract: Finite Element Analysis (FEA) serves as the cornerstone of modern engineering design.
Read next because VFEAgent: A Multimodal Agent Framework for End-to-End Automated Finite Element Analysis overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, rect, correct, eval, line, rate, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.28978v1 Announce Type: new Abstract: Finite Element Analysis (FEA) serves as the cornerstone of modern engineering design. However, its workflow is inherently complex and relies heavily on domain expertise. Although recent efforts have integrated Large Language Models (LLMs) into FEA, existing approaches face limitations in handling multimodal inputs and executing complex tasks. To address these limitations, we propose VFEAgent, an end-to-end multi-agent system designed to automate FEA modeling and simulation directly from input images and problem descriptions. Our methodology integrates two core components: (1) a multimodal vision-language multi-agent pipeline that employs ReAct-driven reasoning to extract structured FEA specifications from heterogeneous inputs and (2) a verification-first code synthesis framework, incorporating robust self-debugging and fallback mechanisms to ensure executability and physical validity. We systematically evaluated the system across various engineering mechanics scenarios. The results demonstrate that VFEAgent achieves a high success rate in generating complete and physically valid simulations, outperforming LLM-based baseline methods in reliability and correctness. These findings validate the feasibility of automating the complete FEA workflow, highlighting the framework's potential to liberate engineers from tedious manual analysis.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.28965unread
Frontier LLM-based agents can overcome the ontology curation bottleneck for natural phenotypes
James P. Balhoff, Hilmar Lapp · 2026-05-29
arXiv:2605. 28965v1 Announce Type: new Abstract: Linking free-text phenotype descriptions to ontology terms, typically referred to as phenotype annotation, is essential for the cross-study integration of comparative morphological data.
Read next because Frontier LLM-based agents can overcome the ontology curation bottleneck for natural phenotypes overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, eval, source, anth, project, trained. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.28965v1 Announce Type: new Abstract: Linking free-text phenotype descriptions to ontology terms, typically referred to as phenotype annotation, is essential for the cross-study integration of comparative morphological data. This labor intensive process has heavily relied on highly trained human experts, which makes it challenging to scale and thus a key bottleneck. Dahdul et al. (2018) established a Gold Standard (GS) of Entity-Quality (EQ) annotations across seven phylogenetic studies and used it to evaluate three human curators and the Semantic CharaParser NLP tool with ontology-based semantic similarity metrics; they reported that machine-human consistency was significantly lower than inter-curator (human-human) consistency. Here we revisit that benchmark with five frontier hosted LLMs from Anthropic and OpenAI, each operating as an "agentic curator" within a self-contained workspace that supplies the source publication PDF, the same annotation guide used by the original human curators, the four project ontologies (UBERON, PATO, BSPO, GO), and a validation script. Evaluated against the same Gold Standard, every agent fell within the range of inter-curator variability of the three trained human biocurators of the original study; the best performing agents approached but did not reach the best performing human curator. Agents substantially outperformed Semantic CharaParser on all four metrics.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.28902unread
Orthogonal Concept Erasure for Diffusion Models
Yuhao Sun, Lingyun Yu, Haoxiang Xu, Fengyuan Miao, Zhuoer Xu, Hongtao Xie · 2026-05-29
arXiv:2605. 28902v1 Announce Type: new Abstract: Concept erasure has emerged as a promising approach to mitigate undesired or unsafe content in diffusion models, yet existing methods still face significant limitations.
Read next because Orthogonal Concept Erasure for Diffusion Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, rect, rate, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.28902v1 Announce Type: new Abstract: Concept erasure has emerged as a promising approach to mitigate undesired or unsafe content in diffusion models, yet existing methods still face significant limitations. While training-based methods are effective, their high computational cost limits scalability. Editing-based methods are more efficient and deployment-friendly, yet they struggle to simultaneously achieve precise concept erasure and preserve overall generative capacity. We identify this core limitation of the editing-based methods as reliance on additive parameter updates. Our empirical analysis reveals that concept semantics primarily depend on neuron direction rather than neuron magnitude, while overall generative capacity relies on the angular geometry of neurons. As additive updates inherently entangle direction, magnitude, and angular geometry, they inevitably introduce unintended interference between concept erasure and overall generation performance. To address this, we propose Orthogonal Concept Erasure (OCE), which reformulates editing-based erasure as multiplicative parameter updates from a geometric perspective. Specifically, OCE applies layer-wise orthogonal transformations derived from a closed-form solution to the parameters, enabling precise concept erasure while preserving the neuron magnitude and angular geometry. Furthermore, to address conflicting constraints in multi-concept erasure, OCE introduces a subspace-level objective with structured subspace manipulation, yielding a more effective and scalable erasure. Extensive experiments on single- and multi-concept erasure demonstrate that OCE outperforms existing methods in concept erasure and non-target preservation, erasing up to 100 concepts in 4.3 s. Code: https://github.com/HansSunY/OCE.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.28864unread
The Cognitive Categorical Transformer: Category-Theoretic Inductive Biases for Language Modeling
Al Kari · 2026-05-29
arXiv:2605. 28864v1 Announce Type: new Abstract: The Cognitive Categorical Transformer (CCT) is a 306M-parameter architecture that augments a pretrained GPT-2 Small backbone with cognitively grounded components derived from category theory and several inspirations from cognitive science.
Read next because The Cognitive Categorical Transformer: Category-Theoretic Inductive Biases for Language Modeling overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, under, line, compare, alone, full, trained, language. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.28864v1 Announce Type: new Abstract: The Cognitive Categorical Transformer (CCT) is a 306M-parameter architecture that augments a pretrained GPT-2 Small backbone with cognitively grounded components derived from category theory and several inspirations from cognitive science. Under a matched-step protocol (215,000 optimizer steps, matched data, matched optimizer and schedule) on WikiText-103, CCT reaches 21.27 validation perplexity, compared with 24.19 for an identically fine-tuned GPT-2 Small baseline. The architecture therefore contributes a 2.92 PPL (12% relative) reduction beyond what in-domain fine-tuning alone provides. A retrain-from-scratch ablation that holds GT-Full simplicial message passing bypassed across the entire seven-phase activation schedule reaches 23.72 PPL, localizing 84% of the architectural improvement (2.45 of 2.92 PPL) to GT-Full. We present the first ablation-validated evidence that simplicial message passing improves language-model perplexity at the 306M-parameter scale on WikiText-103. Published GPT-2 Large reaches 22.05 zero-shot PPL on WikiText-103 with 6.2x more parameters than GPT-2 Small; this paper treats that number as an external published reference, not as the architectural benchmark. Three negative results on consistency-style categorical priors (sheaf smoothing, adjunction round-trip, curvature regularization) and the joint structural-prior result for GT-Full and PrecisionWeightedPP together support an empirical pattern termed the *structure/consistency distinction*, in which categorical priors that add new topology improve language modeling and those that enforce a consistency identity do not.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias, negative, benchmark.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.28855unread
Behavior-Aware Auxiliary Corrections for Off-Policy Temporal-Difference Prediction
Xingguo Chen, Zhiang He, Yuchen Shen, Shangdong Yang, Chao Li, Guang Yang, Wenhao Wang · 2026-05-29
arXiv:2605. 28855v1 Announce Type: new Abstract: Temporal-difference learning with function approximation can be unstable under off-policy sampling.
Read next because Behavior-Aware Auxiliary Corrections for Off-Policy Temporal-Difference Prediction overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: rect, under, correct, line, rate, compare, chain, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.28855v1 Announce Type: new Abstract: Temporal-difference learning with function approximation can be unstable under off-policy sampling. TDC stabilizes off-policy TD through an auxiliary covariance correction, and TDRC further regularizes this correction in a single-timescale recursion. This paper studies a behavior-aware replacement of the auxiliary covariance geometry in the linear prediction setting, which is the standard local model for understanding the feature-space dynamics of value-function approximation. We first replace the TDC auxiliary matrix (C) by the behavior Bellman matrix (A_\mu), yielding BA-TDC, and then regularize the same behavior-aware equation to obtain BA-TDRC. This two-step construction separates the contribution of behavior-aware geometry from the contribution of regularization. The linear analysis also provides a tractable model for an auxiliary-geometry design question that arises in neural-network value approximation, where feature covariances and temporal transition matrices jointly shape the last-layer correction dynamics. We give a finite-state mean-system formulation, prove fixed-point preservation and almost-sure convergence under a Hurwitz stability condition on the instantiated mean system, and compare deterministic mean rates through the spectral radius of the exact linear error recursion. Experiments on the two-state counterexample, Baird's counterexample, Random Walk, and Boyan Chain show that the behavior-aware replacement can be highly beneficial by itself on some tasks, but that regularization is necessary for robust performance across harder settings.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses counterexample.
source ↗open
score 100arxiv cs.AI (Artificial Intelligence)arxiv:2605.28849unread
Behavior-Induced Mirror-Prox Temporal-Difference Learning for Faster Off-Policy Prediction
Xingguo Chen, Yuchen Shen, Shangdong Yang, Chao Li, Guang Yang, Wenhao Wang · 2026-05-29
arXiv:2605. 28849v1 Announce Type: new Abstract: Gradient temporal-difference methods provide stable off-policy prediction with linear function approximation, but their practical performance is strongly affected by the geometry induced by the auxiliary-variable metric.
Read next because Behavior-Induced Mirror-Prox Temporal-Difference Learning for Faster Off-Policy Prediction overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rect, under, correct, line, rate, project, chain. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2605.28849v1 Announce Type: new Abstract: Gradient temporal-difference methods provide stable off-policy prediction with linear function approximation, but their practical performance is strongly affected by the geometry induced by the auxiliary-variable metric. Existing Mirror-Prox TD methods typically use the feature covariance metric, whereas hybrid TD methods suggest that behavior-policy transition information can provide a more informative update geometry. This paper proposes a behavior-induced Mirror-Prox temporal-difference method, called STHTD-MP, which replaces the covariance metric in the primal-dual saddle-point formulation with the symmetric part of the behavior-policy Bellman matrix. The method keeps a single learning rate for the primal and auxiliary variables and applies a Mirror-Prox prediction-correction step to the resulting hybrid saddle-point operator. We provide a formal convergence analysis for fixed-policy linear prediction under standard stochastic approximation assumptions: the behavior-induced metric is positive definite, the joint mean system is Hurwitz, boundedness follows from a Lyapunov argument, and the stochastic recursion converges by the ODE method. We further derive projected-oracle ergodic gap bounds and an exact mean-operator comparison with GTD2-MP based on the spectral radius of the deterministic Mirror-Prox error matrix. The analysis shows that STHTD-MP can have a smaller mean contraction factor than GTD2-MP when the behavior-induced metric improves the saddle-point geometry. Exact numerical mean-operator analysis on two-state, Random Walk, and Boyan Chain benchmarks supports this condition, while Baird's counterexample is identified as a singular boundary case where the strict assumptions fail.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses counterexample, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27767unread
UniMaia: Steering Chess Policies with Language for Human-like Play
Sherman Siu (University of Waterloo), Lesley Istead (University of Waterloo) · 2026-05-29
arXiv:2605. 27767v1 Announce Type: new Abstract: Recent advances in large language models have enabled natural language to serve as a flexible interface for controlling complex systems, but often at the cost of large-scale multimodal training or weakened domain-specific inductive biases.
Read next because UniMaia: Steering Chess Policies with Language for Human-like Play overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, strong, text, eval, line, rate, control, without. Source: arxiv cs.CL (NLP).
arXiv:2605.27767v1 Announce Type: new Abstract: Recent advances in large language models have enabled natural language to serve as a flexible interface for controlling complex systems, but often at the cost of large-scale multimodal training or weakened domain-specific inductive biases. In structured decision-making domains such as chess, specialized policy networks achieve strong performance but lack semantic controllability, while prompt-conditioned language models are more flexible yet typically exhibit weaker domain grounding. We propose $\textbf{UniMaia}$, a framework for prompt-conditioned policy modulation that adapts a frozen Lc0-based chess policy network using a parameter-efficient text encoder and a ControlNet-style conditioning mechanism. UniMaia enables semantic control over gameplay, including opening selection and player strength, while preserving the pretrained policy representations. We further introduce $\textbf{UniMaia-Aux}$, which incorporates auxiliary temporal conditioning and behavioral prediction objectives. To support this work, we construct a large-scale metadata-augmented Lichess dataset, develop a semi-automated prompt-generation pipeline, and introduce benchmarks spanning both prompt-conditioned and metadata-conditioned settings. UniMaia achieves state-of-the-art expected accuracy on several prompt-conditioned benchmarks and competitive top-move accuracy on general instruction-following tasks, while remaining competitive with dedicated metadata-conditioned approaches on human move prediction benchmarks. UniMaia-Aux further improves expected accuracy and behavioral modeling across several evaluation settings, with modest trade-offs in top-move accuracy. Overall, our results demonstrate that prompt-conditioned control of domain-specific policy networks is feasible without end-to-end multimodal training, while highlighting trade-offs between controllability and predictive performance.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias, evaluation, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27750unread
Reading or Guessing? Visual Grounding Failures of Vision-Language Models for OCR in Ancient Greek Editions
Antonia Karamolegkou, Nicolas Angleraud, Beno\^it Sagot, Thibault Cl\'erice · 2026-05-29
arXiv:2605. 27750v1 Announce Type: new Abstract: Recent work has shown that Vision-Language Models (VLMs) used for optical character recognition (OCR) can generate plausible but visually unsupported text, suggesting reliance on language priors.
Read next because Reading or Guessing? Visual Grounding Failures of Vision-Language Models for OCR in Ancient Greek Editions overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, text, rect, under, correct, wrong, eval, source. Source: arxiv cs.CL (NLP).
arXiv:2605.27750v1 Announce Type: new Abstract: Recent work has shown that Vision-Language Models (VLMs) used for optical character recognition (OCR) can generate plausible but visually unsupported text, suggesting reliance on language priors. Comparing open-weight VLMs with traditional OCR baselines on low-resource Ancient Greek critical editions, we show that VLM errors often remain fluent even when wrong, producing plausible Greek substitutions where traditional engines produce local recognition noise. To analyze visual evidence during decoding, we introduce controlled image perturbations and token-level grounding measures based on conditional versus image-free decoding distributions. Under character-level perturbations, VLMs diverge sharply from the perturbed ground truth while traditional OCR remains comparatively faithful; however, token-level analysis shows that prior reliance is model-specific: in an OCR-specialist model, fluent lexical errors are produced with little reliance on the image, whereas general-purpose VLMs remain conditioned on the visual input even when wrong. Decode-time interventions fail to reliably restore grounding, while post-OCR language-model correction improves several systems only by repairing text after generation. Our results extend prior evidence of OCR language-prior reliance to low-resource historical documents and a broader set of models, showing that fluent output is not necessarily visually grounded and motivating interpretability-driven evaluation beyond aggregate accuracy.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures, evaluation.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27741unread
Escape the Language Prior: Mitigating Late-Stage Modality Collapse in Audio Reasoning via Modality-Aware Policy Optimization
Cihan Xiao, Yiwen Shao, Chenxing Li, Xiang He, Zhenwen Liang, Steve Yves, Sanjeev Khudanpur, Liefeng Bo · 2026-05-29
arXiv:2605. 27741v1 Announce Type: new Abstract: Audio and omni-modal large language models exhibit impressive cross-modal reasoning capabilities.
Read next because Escape the Language Prior: Mitigating Late-Stage Modality Collapse in Audio Reasoning via Modality-Aware Policy Optimization overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, eval, source, token, rate, chain, stage, language. Source: arxiv cs.CL (NLP).
arXiv:2605.27741v1 Announce Type: new Abstract: Audio and omni-modal large language models exhibit impressive cross-modal reasoning capabilities. However, applying standard reinforcement learning post-training algorithms to these models exposes a critical structural vulnerability: methods like GRPO apply uniform policy gradients across all tokens, ignoring their unequal dependence on the non-text source modality. This exacerbates late-stage modality collapse during extended chain-of-thought generation, where models progressively abandon the primary source signal in favor of compressed textual priors, leading to confident but ungrounded hallucinations. To address this, we introduce Modality-Aware Policy Optimization (MAPO), a novel dual-branch reinforcement learning framework. First, MAPO dynamically concentrates the policy gradient on modality-critical tokens using a modality relevance mask, which is derived from the cross-modal differential entropy between an audio-ablated reference and the multimodal policy. Second, it integrates an auxiliary attention loss branch that applies a targeted, temporally scaled penalty to the model's internal attention distributions. This ensures the model actively sustains cross-modal grounding deep into the reasoning trace. Evaluations on complex audio reasoning benchmarks demonstrate that MAPO substantially improves long-horizon reasoning fidelity and multimodal instruction following, achieving highly competitive performance and setting new state-of-the-art results on several key benchmarks among open-weight models. By relying strictly on native statistical signals rather than domain-specific inductive biases, MAPO offers a promising foundation for mitigating epistemic collapse across diverse multimodal systems.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias, evaluation, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27740unread
UNIQUE: Universal Top-k Sparse Attention for Training-free Inference and Sparsity-aware Training
Keqi Deng, Shaoshi Ling, Ruchao Fan, Jinyu Li · 2026-05-29
arXiv:2605. 27740v1 Announce Type: new Abstract: Long-context inference in large language models (LLMs) is bottlenecked by the linear growth of the self-attention key-value (KV) cache.
Read next because UNIQUE: Universal Top-k Sparse Attention for Training-free Inference and Sparsity-aware Training overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, soft, line, rate, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27740v1 Announce Type: new Abstract: Long-context inference in large language models (LLMs) is bottlenecked by the linear growth of the self-attention key-value (KV) cache. Top-k sparse attention alleviates this by loading only a small fraction of the KV cache, but accurately and cheaply estimating cache importance, for both training-free use and sparsity-aware training, remains challenging. This paper proposes UNIQUE, a universal top-k sparse attention framework that addresses both requirements and stays consistently effective across LLM modalities. UNIQUE operates at the granularity of KV pages and estimates per-page importance with a simple yet accurate score combining the mean of the page's keys as a representative vector with their standard deviation as an offset term. To further close the train-inference gap, this paper introduces a soft-mask sparsity-aware training scheme that uses the top-k score boundary as a per-query threshold and a sigmoid soft mask around it, requiring neither auxiliary losses nor architectural changes. Experiments on text and speech LLMs show that UNIQUE preserves task performance on long-context benchmarks such as LongBench Pro and on long-form speech recognition, while delivering up to 11.4x attention-kernel speedup over FlashInfer dense attention and at least 5.3x end-to-end decoding speedup over a vLLM-based dense model.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27721unread
UserHarness: Harnessing User Minds for Stronger Agent Theory-of-Mind
Cheng Qian, Jiayu Liu, Heng Ji · 2026-05-29
arXiv:2605. 27721v1 Announce Type: new Abstract: Understanding what a user believes and intends is central to building effective agent assistants.
Read next because UserHarness: Harnessing User Minds for Stronger Agent Theory-of-Mind overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rect, under, eval, assistant, line, without, does. Source: arxiv cs.CL (NLP).
arXiv:2605.27721v1 Announce Type: new Abstract: Understanding what a user believes and intends is central to building effective agent assistants. This ability is often evaluated through Theory-of-Mind (ToM) tasks, where success requires reasoning from the user's perspective. However, many existing approaches address ToM with complex pipelines that model behavior indirectly, without explicitly reconstructing the user's mental state. This misses the core structure of the problem: users act based on their beliefs, which are updated through observations of the environment; beliefs and intentions jointly determine actions, which in turn change the environment; and social reasoning often requires nested beliefs about what others believe or intend. We propose UserHarness, a simple framework that reframes ToM reasoning as explicit user-mind reconstruction. UserHarness decomposes the user's mental state, its relation to the external environment, and the actions that follow from it, enabling agents to track what the user observes, believes, intends, and does. Across five benchmarks, UserHarness reaches up to 95.94% macro accuracy, improving over existing inference methods by more than 15% relative and over the strongest prompt-only harness by about 20% relative. These results suggest that robust user understanding requires reasoning from the roots of the user's mind, positioning user harnessing as a promising foundation for more adaptive future assistants.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27715unread
Beyond Input Understanding: Diagnosing Multilingual Mathematical Reasoning with Directed Acyclic Trace Graphs
Jiaqiao Zhang, Zhoujun Li, Raoyuan Zhao, Jian Lan, Thomas Seidl, Michael A. Hedderich, Hinrich Sch\"utze, Yihong Liu · 2026-05-29
arXiv:2605. 27715v1 Announce Type: new Abstract: Large reasoning models (LRMs) achieve strong mathematical reasoning performance in English, but remain much less reliable in many low- and medium-resource languages.
Read next because Beyond Input Understanding: Diagnosing Multilingual Mathematical Reasoning with Directed Acyclic Trace Graphs overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, rect, under, source, control, test, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27715v1 Announce Type: new Abstract: Large reasoning models (LRMs) achieve strong mathematical reasoning performance in English, but remain much less reliable in many low- and medium-resource languages. This gap is often explained as a failure to understand non-English problem statements. We show that this view is incomplete: even when the problem is given in English, controlling the model's reasoning language can substantially reduce accuracy, suggesting that language also affects reasoning execution itself. To study this effect, we introduce DATG, a Directed Acyclic Trace Graph framework that maps reasoning traces to language-independent mathematical anchors and dependencies. This allows us to align target-language traces with reference DAGs and measure whether they cover required mathematical nodes, respect dependency edges, and avoid harmful mathematical actions. Experiments on the Qwen3 series across 12 languages show that non-English reasoning often suffers from reduced anchor coverage and weaker dependency fidelity, especially in low-resource languages. Motivated by this diagnosis, we propose Loop-Retry and Formula-Retry, two simple test-time controls targeting DATG-exposed failure modes, and show that they consistently improve target-language reasoning performance in low-resource languages.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27709unread
ReverseMath: Answer Inversion for Scalable and Verifiable Mathematical Problem Generation
Raoyuan Zhao, Yihong Liu, Yupei Du, Hinrich Sch\"utze, Michael A. Hedderich · 2026-05-29
arXiv:2605. 27709v1 Announce Type: new Abstract: Mathematical reasoning benchmarks are vital for evaluating large language models (LLMs), but many are static and repeatedly exposed through public evaluation and training pipelines, making it difficult to separate genuine reasoning from memorization.
Read next because ReverseMath: Answer Inversion for Scalable and Verifiable Mathematical Problem Generation overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: rect, correct, eval, source, line, rate, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27709v1 Announce Type: new Abstract: Mathematical reasoning benchmarks are vital for evaluating large language models (LLMs), but many are static and repeatedly exposed through public evaluation and training pipelines, making it difficult to separate genuine reasoning from memorization. Meanwhile, manually constructing new math problems with reliable answers remains costly. We introduce ReverseMath, a scalable method for generating new math problems through answer inversion. Given a problem and its answer, ReverseMath masks a numerical value in the original problem, treats the original answer as a known condition, and rewrites the problem so that the masked value becomes the new answer. The generated problem reverses the original input-output relation, making its answer known by construction. We study ReverseMath for both evaluation and training. For evaluation, paired original/reversed problems reveal substantial behavioral shifts: models sometimes fail on reversed problems and even incorrectly output the original answer, suggesting memorization-like behavior. For training, ReverseMath provides automatically labeled reversed problems as data augmentation for reinforcement learning (RL). Experiments show that including ReverseMath-generated data improves mathematical reasoning performance across multiple benchmarks, demonstrating its value as both an analysis tool and a scalable source of verifiable training data.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses evaluation, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27706unread
Chain-based Adaptive Reconfiguration Over Lattices for Hallucination Reduction
Joan Vendrell Gallart, Solmaz Kia, Russell Bent, Michael Grosskopf · 2026-05-29
arXiv:2605. 27706v1 Announce Type: new Abstract: We introduce CAROL (Chain-based Adaptive Reconfiguration Over Lattices), a probabilistic framework for test-time hallucination reduction in large language models.
Read next because Chain-based Adaptive Reconfiguration Over Lattices for Hallucination Reduction overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, eval, token, line, rate, compare, chain, test. Source: arxiv cs.CL (NLP).
arXiv:2605.27706v1 Announce Type: new Abstract: We introduce CAROL (Chain-based Adaptive Reconfiguration Over Lattices), a probabilistic framework for test-time hallucination reduction in large language models. Rather than relying on token-level uncertainty, CAROL defines a semantic uncertainty measure based on the consistency between generated responses and a trusted context, inducing a string-submodular objective over a lattice of textual sequences. This formulation enables hallucination mitigation to be cast as a Markov chain accept-reject process with provable convergence and near-optimality guarantees, allowing the model to iteratively refine outputs toward semantic consistency. By operating at the level of meaning, CAROL unifies hallucination detection and mitigation within a single framework. Empirical results on question answering and multi-agent reasoning benchmarks show that CAROL significantly reduces hallucinations and improves reliability and interpretability compared to likelihood-based and retrieval-augmented baselines, while maintaining competitive computational efficiency.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27690unread
TRACES: Proactive Safety Auditing for Multi-Turn LLM Agents via Trajectory-State Modeling
Jiaqian Li, Yanshu Li, Boxuan Zhang, Ruixiang Tang, Kuan-Hao Huang · 2026-05-29
arXiv:2605. 27690v1 Announce Type: new Abstract: LLM agents increasingly operate through multi-turn tool use and environment interaction, where safety risks often emerge from intermediate steps long before they surface in the final outcome.
Read next because TRACES: Proactive Safety Auditing for Multi-Turn LLM Agents via Trajectory-State Modeling overlaps with clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "Follow-up to #354: cascading chunk-binding — does A→B, B→C, C→D propagate the full chain on a recipient trained only to emit A?". Matching terms: prefix, rate, full, trained, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27690v1 Announce Type: new Abstract: LLM agents increasingly operate through multi-turn tool use and environment interaction, where safety risks often emerge from intermediate steps long before they surface in the final outcome. Reactive auditing is therefore insufficient: post-hoc diagnosis frequently misses the chance to flag risks while they are unfolding. We propose TRACES, a representation-based proactive auditor that learns prefix-level trajectory risk states from the hidden representations of an observer LLM. TRACES induces latent mechanism features from step representations and models their temporal evolution to estimate whether a partial trajectory is drifting toward unsafe behavior. To sidestep the cost and ambiguity of step-level risk annotation, TRACES is trained with weak trajectory-level supervision while still producing dense prefix-level risk estimates. Across multiple agent safety benchmarks, TRACES improves both full-trajectory safety prediction and proactive risk discrimination. Our analyses further suggest that these risk states can help train a safer agent, highlighting the broader potential of proactive auditing for long-horizon agent safety.
Potential threat/caveat for clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27654unread
Cultural Fidelity in English-to-Hindi Translation: A Preservation-Fluency Frontier for Gender Recoverability
Samyak Savi, Chavi Gupta, Shreyas Gantayet, Tanay Sodha, Dhruv Kumar · 2026-05-29
arXiv:2605. 27654v1 Announce Type: new Abstract: Generative translation systems are cultural technologies because they decide how socially meaningful cues are rendered within culturally specific grammatical systems.
Read next because Cultural Fidelity in English-to-Hindi Translation: A Preservation-Fluency Frontier for Gender Recoverability overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, eval, source, candidates, candidate. Source: arxiv cs.CL (NLP).
arXiv:2605.27654v1 Announce Type: new Abstract: Generative translation systems are cultural technologies because they decide how socially meaningful cues are rendered within culturally specific grammatical systems. We study one concrete notion of successful cultural translation: when an English source explicitly encodes gender, an English-to-Hindi translation should preserve the recoverability of that cue unless the source itself is ambiguous. We evaluate this criterion on a 37,345-instance benchmark spanning twelve categories and show that five systems frequently erase gender through ergative and honorific constructions. We then introduce two mechanism-aware inference-time interventions. The first, the Source-Aware Reranker (SAR), prefers candidates that avoid gender-neutralizing syntax. The second, the Phenomenon-Aware Reranker (PAR), preserves gender through targeted lexical marking even when ergative syntax remains. Across GPT-4o-mini and Sarvam, PAR improves target-subset accuracy from 11.07% to 54.47% and from 15.99% to 49.66%, respectively. Human evaluation shows that PAR increases gender preservation from 10.3% to 81.3%, but reduces mean fluency from 4.36 to 3.37. These findings place the two interventions on a preservation and fluency frontier rather than supporting a single dominant solution, and show how culturally situated generation can require explicit tradeoffs among fidelity, fluency, and stylistic naturalness.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27649unread
Disentangling Language Roles in Multilingual LLM Task Execution
Qishi Zhan, Minxuan Hu, Seoyeon Jang, Lei Zhao, Ziheng Chen, Man Liang, Xinyue Xiang, Jiaxin Liu, Guansu Wang, Liang He · 2026-05-29
arXiv:2605. 27649v1 Announce Type: new Abstract: Multilingual LLMs are increasingly used when instruction, source content, and required response languages do not coincide.
Read next because Disentangling Language Roles in Multilingual LLM Task Execution overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, rect, correct, eval, source, rate, extraction, control. Source: arxiv cs.CL (NLP).
arXiv:2605.27649v1 Announce Type: new Abstract: Multilingual LLMs are increasingly used when instruction, source content, and required response languages do not coincide. Existing benchmarks have expanded multilingual instruction-following evaluation, but they rarely isolate these three roles within a fully crossed design. We introduce MTM-Bench, a controlled benchmark for language-conditioned task execution in which each instance is defined by a triplet $(L_{\text{instr}}, L_{\text{content}}, L_{\text{resp}})$. Across English, Spanish, and Chinese, MTM-Bench enumerates all 27 triplets and contains 2{,}430 instances per model across semantic reversal, final-state extraction, and language purity with update realization. We evaluate 20 frontier and open-weight LLMs using decomposed metrics for semantic correctness, target-language adherence, constraint satisfaction, contamination ratio, and joint success, with scoring validated by a targeted human audit. The fully crossed design reveals that degradation is organized by the role a language occupies in the task structure, not merely by mismatch count. The response-language role is the dominant axis of variation, and a single response-slot mismatch accounts for most degradation. The response-only and full-mismatch comparison suggests that mismatch count is not a monotonic predictor of difficulty, with model-level ordering varying across systems. Task families fail through distinct channels, showing that semantic correctness alone does not capture reliable multilingual task execution.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27636unread
Simorgh at SemEval-2026 task 7: Region-Aware Hybrid Retrieval for Low-Resource Cultural Reasoning in Multilingual Question Answering
Hadi Bayrami Asl Tekanlou, Mahdi Bakhtiyarzadeh, Jafar Razmara · 2026-05-29
arXiv:2605. 27636v1 Announce Type: new Abstract: Although Large Language Models (LLMs) demonstrate excellent capabilities and performance for general reasoning tasks within the general public domain, they may face challenges with culturally grounded knowledge within languages with limited digital and textual data.
Read next because Simorgh at SemEval-2026 task 7: Region-Aware Hybrid Retrieval for Low-Resource Cultural Reasoning in Multilingual Question Answering overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, eval, source, rate, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27636v1 Announce Type: new Abstract: Although Large Language Models (LLMs) demonstrate excellent capabilities and performance for general reasoning tasks within the general public domain, they may face challenges with culturally grounded knowledge within languages with limited digital and textual data. In this paper, we investigate culturally grounded multiple-choice question answering with the BLEnD benchmark, which consists of a multilingual corpus of 30 languages and covers various socio-cultural domains, such as cuisine, sports, family, etc. We propose a region-aware hybrid retrieval approach that combines BM25 lexical matching and dense semantic similarity with regional weighting heuristics to improve the relevance of the answer. The retrieved documents are used to construct a structured prompt for the Qwen3-14B quantized model with logit-based deterministic answer selection. The experimental results show improvements to cross-lingual stability with the hybrid retrieval approach over pure parametric inference for culturally grounded question answering. However, there are still notable performance gaps between languages with more and less training data. This shows that the limitations of the retrieval augmentation approach are not entirely overcome by the training data imbalance problem.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27596unread
Can Hallucinations Be Useful? Solving Multi-Hop Questions With SLMs By Chaining System-I/II Reasoning
Saptarshi Sengupta, Suhang Wang · 2026-05-29
arXiv:2605. 27596v1 Announce Type: new Abstract: Recently, there has been increased interest in Small Language Models (SLMs), which are fast, show good performance, and have lower hardware demands than large language models (LLMs).
Read next because Can Hallucinations Be Useful? Solving Multi-Hop Questions With SLMs By Chaining System-I/II Reasoning overlaps with clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: good, eval, source, rate, chain, position, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27596v1 Announce Type: new Abstract: Recently, there has been increased interest in Small Language Models (SLMs), which are fast, show good performance, and have lower hardware demands than large language models (LLMs). However, SLMs hallucinate more frequently than LLMs, impacting their ability to solve complex multi-step reasoning problems as early mistakes cascade to the final response. To address this, existing works think-first followed by iterative retrieval to reduce hallucination. We argue that the think-first strategy is not always necessary as we find that: (i) SLMs are often accurately confident in their initial answer and, (ii) hallucinations can actually be beneficial for honing in on the true answer. As such, we position our work as an inversion of this strategy, i.e., answer first-reason later. We propose a cognitively-inspired framework where the model is first allowed to quickly answer the question (System-I (zero-shot)) and then resorts to deeper thinking (System-II) based on evidence retrieved from a knowledge source using the initial hypothesis. By combining System-I and System-II style thinking, we show that our method can outperform prior work that takes the traditional think-first route on various multi-step question-answering benchmarks.
Potential threat/caveat for clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27564unread
The Future of Facts: Tracing the Factual Generation-Verification Gap
Tim R. Davidson, Anja Surina, Caglar Gulcehre · 2026-05-29
arXiv:2605. 27564v1 Announce Type: new Abstract: Language models are becoming the default interface to factual knowledge, yet they often verify outputs more reliably than they generate them.
Read next because The Future of Facts: Tracing the Factual Generation-Verification Gap overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)". Matching terms: rect, under, correct, source, rate, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27564v1 Announce Type: new Abstract: Language models are becoming the default interface to factual knowledge, yet they often verify outputs more reliably than they generate them. This generation-verification gap (GV-gap) underlies many recent advances in self-improvement and reasoning, but its dynamics on factual knowledge specifically remain poorly understood. We focus on the training mechanisms underlying factual GV-gaps, distinguishing them from their computational and aesthetic counterparts. We trace generation and verification capabilities through three training phases (acquisition, continual learning, and updating) across four open-source model families at two scales each. Three findings recur across models: (i) verification is consistently learned before generation; (ii) verification is more robust to continual learning than generation; and (iii) factual updates can leave models in a "multi-verse" state, simultaneously verifying both old and new answers as correct. Natural experiments on frontier models reproduce these dynamics at scale and reveal residual verification biases on well-covered facts.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses bias.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27545unread
PAST2HARM: A Simple Adaptive Past Tense Attack for Jailbreaking Multimodal AI
Snehasis Mukhopadhyay · 2026-05-29
arXiv:2605. 27545v1 Announce Type: new Abstract: Jailbreak attacks on multimodal AI systems remain underexplored, even though unsafe image generation can have more severe consequences than unsafe text and current defenses are relatively immature.
Read next because PAST2HARM: A Simple Adaptive Past Tense Attack for Jailbreaking Multimodal AI overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, text, under, alignment, eval, source, rate, language. Source: arxiv cs.CL (NLP).
arXiv:2605.27545v1 Announce Type: new Abstract: Jailbreak attacks on multimodal AI systems remain underexplored, even though unsafe image generation can have more severe consequences than unsafe text and current defenses are relatively immature. We introduce PAST2HARM, a simple yet effective adaptive jailbreak framework that bypasses refusal training in state of the art multimodal text to image models. Building on prior findings that past tense reformulations can evade safeguards, PAST2HARM systematically exploits this vulnerability in multimodal generative AI. We characterize the attack along two dimensions. First, breadth: through temporal deepening, the framework incrementally strengthens historical anchoring and archival cues, eroding refusal boundaries across models with varying alignment strength. Second, depth: via iterative escalation after initial compliance, we probe the upper bound of harmful generation, measuring severity using a scalar severity jailbreak metric evaluated by a language model acting as a judge. We find that mid conversation turns form peak vulnerability windows, where harmfulness increases before plateauing and eventually undergoing semantic inversion. We evaluate PAST2HARM on three models Gemini Nano Banana Pro, GPT Image 2, and SD XL achieving attack success rates of 83 percent, 67 percent, and 100 percent in a black box, gradient free setting. Adversarial prompts also transfer across models, with cross model success rates above 50 percent. The attack elicits diverse harmful outputs, including explicit sexual content, political disinformation, historical denial narratives, hate speech, and self harm glorification. We further release a curated benchmark of prompts, reformulations, and outputs as a resource for red teaming and alignment. Our results expose fundamental brittleness in current safeguards and highlight the need for stronger multimodal safety training.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses adversarial, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27393unread
StoryMI: Steerable Multi-Agent Therapeutic Dialogue Generation
Qingyu Meng, Min Chen, Dingming Liu, Yifan Mo, Yue Su, Xin Sun, Koen Hindriks, Jiahuan Pei · 2026-05-29
arXiv:2605. 27393v1 Announce Type: new Abstract: Large language models (LLMs) can generate fluent dialogue, but prior works lack situational grounding, dynamic strategy control, and evaluation aligned with clinical standards in motivational interviewing (MI).
Read next because StoryMI: Steerable Multi-Agent Therapeutic Dialogue Generation overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, text, eval, source, rate, control, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27393v1 Announce Type: new Abstract: Large language models (LLMs) can generate fluent dialogue, but prior works lack situational grounding, dynamic strategy control, and evaluation aligned with clinical standards in motivational interviewing (MI). We introduce StoryMI, a multi-LLM agent framework for controllable MI dialogue generation, where questionnaire-based client profiles are expanded into situational stories that provide narrative context for the dialogue. Therapist and client agents generate MI-coded utterances guided by MI codes selected by the interaction agent, while an interaction agent dynamically coordinates exchanges to control MI strategies during a multi-turn conversation. We propose a two-level evaluation protocol: lexical metrics and MI-specific measures of macro-level counseling strategies, alongside LLM-as-judge and human expert assessments. We construct a dataset of 6K simulated MI dialogues grounded in 1K questionnaire-story pairs, covering 12 MI codes and 13 symptom domains, and benchmark six open- and closed-source LLMs. Our results show that situational grounding and macro-level control can improve MI adherence and clinical plausibility, demonstrating the effectiveness of a structured multi-agent workflow for psychotherapy dialogue generation. We provide code and data for reproducibility.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27390unread
EvoSpec: Evolving Speculative Decoding via Real-Time Vocabulary and Parameter AdaptationTarget
Shuyu Zhang, Lingfeng Pan, Qicheng Wang, Yaqi Shi, Yueyang Tan, Ruyu Yan, Jiaqi Chen, Lixing Du, Lu Wang · 2026-05-29
arXiv:2605. 27390v2 Announce Type: new Abstract: Speculative decoding accelerates Large Language Model inference via a draft-then-verify paradigm, yet the output projection layer becomes a bottleneck as vocabulary sizes scale.
Read next because EvoSpec: Evolving Speculative Decoding via Real-Time Vocabulary and Parameter Adaptation overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, alignment, distributional, eval, token, line, rate, project. Source: arxiv cs.CL (NLP).
arXiv:2605.27390v2 Announce Type: new Abstract: Speculative decoding accelerates Large Language Model inference via a draft-then-verify paradigm, yet the output projection layer becomes a bottleneck as vocabulary sizes scale. While existing static pruning methods effectively reduce this overhead, they suffer from precipitous drops in acceptance rate in specialized domains or topic-switching scenarios due to their inability to capture dynamic distribution shifts. To address this, we introduce EvoSpec, a framework that enables real-time evolution of the draft model through dynamic vocabulary and parameter adaptation. Unlike static or purely retrieval-based approaches, EvoSpec employs a context-aware mechanism that retrieves critical long-tail tokens via efficient semantic and statistical indexing. Furthermore, we propose a lightweight online alignment strategy utilizing curriculum learning to continually minimize the distributional gap between the draft and target models. Extensive evaluations across specialized domains (coding, law, and medicine) confirm that EvoSpec overcomes the limitations of static baselines. On EAGLE-3, it achieves a 1.13x speedup in these settings over the state-of-the-art static baseline FR-Spec, with 27\% lower memory overhead than standard online adaptation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations, evaluation.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27388unread
Modeling Community Attitude through Reaction Tone: A Human-AI Collaborative Framework for Evaluating LLM Alignment with Linguistic Behaviors in Online Communities
Nuan Wen, Xuezhe Ma · 2026-05-29
arXiv:2605. 27388v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly utilized as proxies for computational social analysis; yet, their ability to faithfully represent the "thick descriptions" (Geertz, 1973) of human communities remains a critical challenge.
Read next because Modeling Community Attitude through Reaction Tone: A Human-AI Collaborative Framework for Evaluating LLM Alignment with Linguistic Behaviors in Online Communities overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: under, alignment, eval, line, rate, full, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27388v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly utilized as proxies for computational social analysis; yet, their ability to faithfully represent the "thick descriptions" (Geertz, 1973) of human communities remains a critical challenge. Current evaluations often reduce social identity to static labels, sidelining how real-world groups navigate social shifts. To bridge this gap, we introduce CARE (Community-Aware Reaction Evaluation), a reaction-centered framework that benchmarks LLM-simulated discourse against the authentic, event-contingent responses of distinct communities to real-world news. By characterizing a fine-grained spectrum of illocutionary tones and the underlying attitudes they manifest--validated through human-AI collaboration--our diagnosis reveals a persistent "realism gap": steering LLMs with explicit community prompts fails to inherently improve simulation fidelity. Analysis further identifies divergent behavioral signatures among frontier models, suggesting that current alignment strategies remain insufficient for capturing the sociolinguistic dynamics of online groups.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses evaluation, benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27380unread
BioELX: Cross-lingual Biomedical Entity Linking via Alias-based Retrieval and LLM Ranking
Yi Wang, Corina Dima, Liangyu Zhong, Steffen Staab · 2026-05-29
arXiv:2605. 27380v1 Announce Type: new Abstract: Cross-lingual biomedical entity linking (BEL) maps mentions in any language to unique identifiers in a biomedical knowledge base (KB), supporting clinical and biomedical NLP applications.
Read next because BioELX: Cross-lingual Biomedical Entity Linking via Alias-based Retrieval and LLM Ranking overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, text, eval, source, trained, stage, candidate, language. Source: arxiv cs.CL (NLP).
arXiv:2605.27380v1 Announce Type: new Abstract: Cross-lingual biomedical entity linking (BEL) maps mentions in any language to unique identifiers in a biomedical knowledge base (KB), supporting clinical and biomedical NLP applications. However, expert-annotated training data for BEL are costly, especially for low-resource languages. Moreover, many cross-lingual BEL systems rely on SapBERT-based retrievers trained on predominantly English aliases in the KB, leading to poor generalization to unseen non-English mentions and limited context-aware disambiguation. We propose BioELX, a two-stage cross-lingual BEL framework that requires no task-specific annotated training corpora. In Stage~1, we enrich SapBERT training with Wikidata-derived multilingual aliases and use the resulting retriever to improve cross-lingual candidate retrieval. In Stage~2, we perform context-aware disambiguation with a pre-trained LLM ranker that jointly considers the mention context and candidate, eliminating the need for supervised training. Experiments on five benchmarks (XL-BEL, EMEA, Patent, WikiMed-DE, and MedMentions) show that BioELX achieves new state-of-the-art performance. It improves average Recall@1 on XL-BEL by +19.2, with especially large gains for low-resource languages, e.g., +21.6 on Turkish, +22.1 on Korean, +30.8 on Thai, and delivers consistent improvements on EMEA (+6.2), Patent (+5.4), and WikiMed-DE (+12.8). Code and resources will be released upon publication.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27378unread
OralAgent: Integrating Reasoning, Tools, and Knowledge for Interactive Dental Image Analysis
Jing Hao, Siyuan Dai, Yongxin Zhang, Yuci Liang, Jiamin Wu, Jiahao Bao, Yuxuan Fan, Zanting Ye, Yanpeng Sun, Xinyu Zhang, Ming Hu, Liang Zhan, James Kit Hon Tsoi, Linlin Shen, Junjun He, Kuo Feng Hung · 2026-05-29
arXiv:2605. 27378v1 Announce Type: new Abstract: Dental image analysis plays a pivotal role in supporting accurate diagnosis and treatment planning in oral healthcare.
Read next because OralAgent: Integrating Reasoning, Tools, and Knowledge for Interactive Dental Image Analysis overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, text, class, eval, source, token, rate, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27378v1 Announce Type: new Abstract: Dental image analysis plays a pivotal role in supporting accurate diagnosis and treatment planning in oral healthcare. Although recent advances have produced dental AI models for specific tasks and individual imaging modalities, their isolated designs limit practical use in real-world clinical workflows. In this paper, we present OralAgent, the first dental-specialized AI agent that unifies multimodal reasoning, tool-based decision-making, and knowledge-grounded retrieval within an end-to-end automated framework. It integrates 22 visual analysis tools and 368 widely-used classical dental textbooks, enabling autonomous reasoning, planning, tool use, knowledge retrieval, and multi-step workflow execution. Furthermore, we introduce OralCorpus, a large-scale, high-quality bilingual textual resource containing 134.8M tokens curated for dental retrieval-augmented generation (RAG). To evaluate models' multidisciplinary dental knowledge, we construct OralQA-ZH, a Chinese multiple-choice question benchmark consisting of 798 items across eleven oral subspecialties. Extensive experiments demonstrate that OralAgent achieves state-of-the-art performance on the MMOral-Uni, MMOral-OPG, and OralQA-ZH benchmarks, highlighting its effectiveness, interpretability, and adaptability in real-world clinical settings. The code and models are publicly available at https://github.com/isjinghao/OralAgent.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27376unread
Unlocking Fine-Grained and Within-Utterance Speaking Style Control in Prompt-Based Text-to-Speech Models
Jaehoon Kang, Yejin Lee, Yoonji Park, Kyuhong Shim · 2026-05-29
arXiv:2605. 27376v1 Announce Type: new Abstract: While prompt-based text-to-speech (TTS) models enable natural language-driven speaking style control, they often provide limited fine-grained control and apply a single global style across an utterance.
Read next because Unlocking Fine-Grained and Within-Utterance Speaking Style Control in Prompt-Based Text-to-Speech Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, strong, text, rect, token, rate, control, language. Source: arxiv cs.CL (NLP).
arXiv:2605.27376v1 Announce Type: new Abstract: While prompt-based text-to-speech (TTS) models enable natural language-driven speaking style control, they often provide limited fine-grained control and apply a single global style across an utterance. This restricts practical use cases that require continuous style attribute interpolation across utterances and time-varying style transitions within a single utterance. In this paper, we propose novel techniques to achieve both capabilities in existing prompt-based TTS models. For inter-utterance style interpolation, we compute direction vectors between contrastive style prompts in the embedding space and perform simple interpolation, enabling smooth transitions between style characteristics. For intra-utterance style transition, we first identify a strong attention bias toward early tokens in autoregressive TTS decoders, causing the initial audio realization to dominate subsequent generation. To mitigate this effect, we introduce KV-cache swapping and sliding-window attention masking. Experiments demonstrate that our proposed inter-utterance interpolation achieves a 99-100% success rate in gender conversion, up to 36 Hz pitch variation, and up to 1.6 syllables-per-second speed change. Our intra-utterance transition maintains a speaker similarity of 0.81-0.91 and achieves perceptual smoothness scores of 3.48-4.48.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias.
source ↗open
score 100arxiv cs.CL (NLP)arxiv:2605.27375unread
LCO: LLM-based Constraint Optimization for Safer Agentic LLMs in Real-world Tasks
Jiayong Wan, Jiawei Chen, Zhaoxia Yin, Liu Shuyuan, Hang Su · 2026-05-29
arXiv:2605. 27375v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly acting as autonomous agents, but their continuous interaction with the environment can lead to in-context reward hacking (ICRH), a phenomenon where LLMs iteratively optimize their behavior to maximize proxy objectives, inadvertently producing harmful side effects.
Read next because LCO: LLM-based Constraint Optimization for Safer Agentic LLMs in Real-world Tasks overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, rate, without, language, model. Source: arxiv cs.CL (NLP).
arXiv:2605.27375v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly acting as autonomous agents, but their continuous interaction with the environment can lead to in-context reward hacking (ICRH), a phenomenon where LLMs iteratively optimize their behavior to maximize proxy objectives, inadvertently producing harmful side effects. Existing defense methods are insufficient to address this risk, as ICRH arises not from adversarial inputs but from the model's own over-optimization. To mitigate this issue, we propose \textbf{LLM-based Constraint Optimization (LCO)}, a framework that effectively reduces ICRH without model fine-tuning. LCO consists of two modules: \textit{self-thought module}, which guides the LLM to proactively deliberate and integrate potential safety constraints before execution; and \textit{evolutionary sampling module}, which employs LLM-based crossover and mutation to constrain the model's actions within a safe solution space while maintaining task performance. Experimental results demonstrate that LCO substantially alleviates ICRH in both output-refine and policy-refine scenarios. In particular, on the tweet engagement optimization task, LCO achieves a 39% reduction in the Toxicity Growth Rate (TGR) on GPT-4, while on the policy optimization benchmark, it reduces the ICRH Occurrence Rate by 15.23%, demonstrating safety improvement without sacrificing task performance.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses adversarial, benchmark.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.29008unread
Causal Intelligence for Constraint-Aware Intervention Design to Induce State Transitions
Zixuan Song, Uwe Mueller, Dimitris V. Manatakis · 2026-05-29
arXiv:2605. 29008v1 Announce Type: new Abstract: Driving a system from one state to another through targeted interventions is a fundamental challenge in science, yet most predictive models offer limited mechanistic insight and no principled framework for decision-making.
Read next because Causal Intelligence for Constraint-Aware Intervention Design to Induce State Transitions overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, distributional, eval, source, rate, trained, another, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.29008v1 Announce Type: new Abstract: Driving a system from one state to another through targeted interventions is a fundamental challenge in science, yet most predictive models offer limited mechanistic insight and no principled framework for decision-making. Here we present COAST (Causally Optimal Actions for State Transitions), a causal-intelligence approach for the in-silico design of constrained interventions that induce user-defined state transitions. Given data characterizing source and target states, COAST learns context-specific causal graphs and structural causal models, attributes observed distributional shifts to mechanism-level causal drivers, and introduces a novel constraint-aware multi-objective optimization formulation that balances transition efficacy, intervention complexity, and target-state stability. The approach is modular and domain-agnostic, integrating feature selection, causal discovery, causal modeling, and intervention identification and evaluation through interchangeable components. Across synthetic benchmarks and real biological datasets, COAST recovers key causal drivers and identifies robust single- and multi-target intervention strategies that achieve desired state transitions, accompanied by transparent mechanistic rationales to guide experimental validation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation, benchmark.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.29005unread
LoRe: Adaptive Interaction-Evaluation Routing with Per-Step Interaction Budgets for Iterative Graph Solvers
Jintao Li, Yong-Yi Wang, Zheng-An Wang, Heng Fan · 2026-05-29
arXiv:2605. 29005v1 Announce Type: new Abstract: Diffusion-based neural solvers for combinatorial optimization repeatedly re-evaluate dense edge/factor interactions, making inference expensive in wall-clock time and often memory-bound at scale.
Read next because LoRe: Adaptive Interaction-Evaluation Routing with Per-Step Interaction Budgets for Iterative Graph Solvers overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, eval, line, full, factor. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.29005v1 Announce Type: new Abstract: Diffusion-based neural solvers for combinatorial optimization repeatedly re-evaluate dense edge/factor interactions, making inference expensive in wall-clock time and often memory-bound at scale. Inspired by the computational methodologies of many-body physics, we introduce LoRe, a training-free, inference-time drop-in wrapper that enforces per-step interaction-evaluation budgeting: at each iteration, it evaluates only a fixed fraction of interactions by dynamically routing computation to high-conflict or high-uncertainty interactions, instead of using a fixed sparsification (e.g., static kNN graphs or static masks). Under fully inclusive end-to-end wall-clock accounting, LoRe substantially improves scalability on the Maximum Independent Set (MIS) problem, extending feasible inference more than $3\times$ beyond the baseline's out-of-memory limit, delivering a $\sim 8\times$ speedup and a $\sim 12\times$ peak-memory reduction, with solution quality preserved in this regime. Demonstrating cross-task generality on the large-scale Traveling Salesperson Problem (TSP) and zero-shot robustness to topology shifts, LoRe achieves a $\sim 15\times$ speedup at $n=1000$ with a $44\times$ memory reduction and competitive tour quality.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses robustness, evaluation.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.29002unread
FedQHD: Closed-Form Function-Space Federated Reinforcement Learning
Yuchen Hou, Yongshan Chen, Zhuowen Zou, Calvin Yeung, Mohsen Imani, Tian Lan, Mahdi Imani · 2026-05-29
arXiv:2605. 29002v1 Announce Type: new Abstract: Federated reinforcement learning enables decentralized agents to collaboratively improve policies or value estimates without exchanging raw trajectories.
Read next because FedQHD: Closed-Form Function-Space Federated Reinforcement Learning overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, alignment, line, rate, project, control, without. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.29002v1 Announce Type: new Abstract: Federated reinforcement learning enables decentralized agents to collaboratively improve policies or value estimates without exchanging raw trajectories. However, FedAvg-style parameter averaging is not function-space consistent: when clients use heterogeneous encoders or even identical nonlinear networks, averaged parameters need not correspond to the weighted average of client value functions in any common function space. We propose FedQHD, a federated Q-learning method using hyperdimensional (random-feature) state encoders with a linear readout, so that Q-functions are nonlinear in state yet linear in trainable parameters. This linear structure enables closed-form aggregation. With a shared encoder, the function-space consensus update coincides exactly with weighted averaging of local readout matrices. With heterogeneous encoders, the server constructs a global teacher by averaging client Q-values on a shared anchor-state set, and each client compiles this teacher into its local representation via a single ridge projection. We formalize the federation gap -- the error incurred when compiling a federated teacher into a heterogeneous client representation -- relative to a client-specific oracle projection. We show that this gap decomposes into subspace misalignment, anchor-set conditioning, and regularization bias. We further identify the anchor-to-dimension ratio $m \geq D_i$ as the well-conditioned regime in which the gap reduces to a multiple of the encoder heterogeneity floor. On four continuous-state, discrete-action control benchmarks, FedQHD matches or outperforms FedAvg-style baselines and distillation-based alternatives while requiring substantially less computation, and the empirical dependence of the federation gap on encoder dimension matches our theoretical analysis.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias, benchmark.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.29001unread
FormInv: A Measurement Protocol for Semantic Invariance in Mathematical Reasoning Benchmarks
Nishal Thomas, Noel Thomas · 2026-05-29
arXiv:2605. 29001v1 Announce Type: new Abstract: A paraphrase-quality audit of MathCheck (ICLR 2025) detected 4 semantically incorrect paraphrases in 129 groups (3.
Read next because FormInv: A Measurement Protocol for Semantic Invariance in Mathematical Reasoning Benchmarks overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)". Matching terms: phrase, phrases, rect, under, correct, eval, rate, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.29001v1 Announce Type: new Abstract: A paraphrase-quality audit of MathCheck (ICLR 2025) detected 4 semantically incorrect paraphrases in 129 groups (3.1%); removing them drops GPT-4o from rank 2 to rank 4 and elevates Claude Haiku and DeepSeek V3 above it; these ranking changes are invisible to any single-model evaluation. Cross-model unanimity found these errors automatically (>= 3/4 models for MathCheck; >= 6/9 for our primary evaluation) for under $10; in our own dataset the same protocol found that 47% of auto-generated connective-variation paraphrases were semantically incorrect. That flaw compounds a deeper measurement gap: Claude Haiku 4.5 achieves 86% accuracy yet SCR=50%, meaning half its theorems are answered differently under semantically equivalent restatements, while aggregate accuracy across 9 models spans only 86-96% yet Semantic Consistency Rates (SCR) span 50-82% -- a 32-point gap invisible to standard benchmarks. Formally, for any target ranking over 9 frontier models there exists a weighting over paraphrase families that realizes it (No-Free-Benchmark corollary), because no model Pareto-dominates all families -- so benchmark designers who select families are implicitly choosing which model wins. FormInv supplies the audit protocol (replicated on external benchmarks at 100% recall), SCR and per-theorem Cochran's Q as primary invariance measures evaluated on 9 models across 366-811 items (on Lean4-verified theorems), and FormInvSelector for regime-aware model selection.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses evaluation, benchmark.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28990unread
Learning Robust and Task-Invariant Functional Representation from fMRI through Siamese Self-Supervised Learning
Jiyao Wang, Peiyu Duan, Nicha C. Dvornek, Lawrence H. Staib, Denis Sukhodolsky, Pamela Ventola, James S. Duncan · 2026-05-29
arXiv:2605. 28990v1 Announce Type: new Abstract: Functional magnetic resonance imaging (fMRI) is a powerful tool for investigating human brain function.
Read next because Learning Robust and Task-Invariant Functional Representation from fMRI through Siamese Self-Supervised Learning overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, class, source, line, rate, full, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28990v1 Announce Type: new Abstract: Functional magnetic resonance imaging (fMRI) is a powerful tool for investigating human brain function. However, the high cost of data acquisition and the inherent subjectivity of psychiatric rating scales often lead to datasets with small sample sizes and variable label quality, especially when targeting a specific neurological condition. Combined with the inherently high dimensionality of fMRI data, these limitations substantially increase the risk of model overfitting. Recent years have seen growing interest in developing fMRI foundation models by combining multiple datasets; however, the computational resources needed for pretraining and fine-tuning are often prohibitive. We show that a lightweight self-supervised framework yields representations that generalize across diverse downstream tasks, outperforming fully supervised baselines and approaching the performance of large-scale models. We introduce BrainSimSiam, a data-efficient self-supervised representation learning framework that leverages positive-only data pairs to learn robust and generalizable features. We demonstrate that the learned representations achieve strong performance across multiple downstream classification and regression tasks, highlighting the potential of BrainSimSiam for data-limited neuroimaging applications.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28983unread
The Hamilton-Jacobi Theory of Deep Learning
Jose Marie Antonio Mi\~noza, Erika Fille T. Legara, Christopher P. Monterola · 2026-05-29
arXiv:2605. 28983v1 Announce Type: new Abstract: In this paper, training a neural network is identified, exactly, as a search through Hamilton--Jacobi initial-value problems: each gradient step selects the initial data of a viscous Hamilton--Jacobi equation whose Hopf--Cole propagator best fits the observations; at inference, the input is the spatial point at which that solution is evaluated and the initial condition is already encoded in the weights.
Read next because The Hamilton-Jacobi Theory of Deep Learning overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, class, under, soft, eval, rate, control. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28983v1 Announce Type: new Abstract: In this paper, training a neural network is identified, exactly, as a search through Hamilton--Jacobi initial-value problems: each gradient step selects the initial data of a viscous Hamilton--Jacobi equation whose Hopf--Cole propagator best fits the observations; at inference, the input is the spatial point at which that solution is evaluated and the initial condition is already encoded in the weights. The correspondence is exact for log-sum-exp layers and structural for broader architectures: residual networks, transformers, and recurrent architectures (RNNs, LSTMs, SSMs) each discretize the same class of Hamilton--Jacobi equations, with architecture-dependent Hamiltonian and viscosity. A single deformation parameter $\varepsilon$ unifies all four perspectives (network, tropical algebra, viscous PDE, convex optimization) in a commutative diagram closed under Lipschitz conditions. Quantitative consequences include: the minimax optimal generalization rate $O(n^{-1/(d+2)})$ for fixed $t$; adversarial robustness controlled by $\varepsilon$; backpropagation as the co-state equation of the Hamiltonian system for residual networks (Pontryagin Maximum Principle); scaling exponents consistent with data intrinsic dimension via PDE quadrature; and a closed-form $O(N)$ influence function (softmax attribution weights $\pi_j$) whose entropy landscape undergoes fold bifurcations as $\varepsilon$ increases, each merging attribution basins.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses robustness, adversarial.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28977unread
Comparing Post-Hoc Explainable AI Methods for Interpreting Black-Box EEG Models in Depression Detection
Antonia \v{S}ar\v{c}evi\'c, Nikolina Frid · 2026-05-29
arXiv:2605. 28977v1 Announce Type: new Abstract: Recent advances in deep learning have enabled increasingly accurate electroencephalography (EEG)-based classification of Major Depressive Disorder (MDD), but the decision-making processes of high-capacity models remain difficult to interpret.
Read next because Comparing Post-Hoc Explainable AI Methods for Interpreting Black-Box EEG Models in Depression Detection overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: marker, class, eval, rate, trained, lora, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28977v1 Announce Type: new Abstract: Recent advances in deep learning have enabled increasingly accurate electroencephalography (EEG)-based classification of Major Depressive Disorder (MDD), but the decision-making processes of high-capacity models remain difficult to interpret. This study investigates multiple post-hoc explainability methods applied to an InceptionTime architecture trained for EEG-based MDD detection. The analysis includes Shapley-based, gradient-based, and perturbation-based attribution approaches: DeepSHAP, Integrated Gradients, GradCAM, Occlusion, and Permutation Feature Importance. Explainability analysis was performed within a subject-level stratified 5-fold cross-validation framework using global attribution aggregation across EEG segments and subjects. The evaluated methods revealed partially convergent attribution patterns, with recurring emphasis on frontal, temporal, and posterior EEG regions, particularly in the right hemisphere. Quantitative comparison demonstrated substantial agreement between gradient- and perturbation-based approaches, while DeepSHAP produced comparatively distinct attribution distributions. At the same time, variability between explainability methods highlighted the influence of methodological assumptions on the resulting explanations. Overall, the results suggest that different post-hoc explainability approaches capture partially overlapping relevance structures in EEG-based deep learning models for depression detection. Although the observed attribution patterns are broadly consistent with several previous EEG studies of MDD, the analysis should be interpreted as exploratory rather than evidence of definitive neurophysiological biomarkers or clinical applicability. The study highlights both the usefulness and limitations of post-hoc explainability for interpreting black-box EEG classifiers in psychiatric applications.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28918unread
When LLM Reward Design Fails: Diagnostic-Driven Refinement for Sparse Structured RL
Youting Wang, Yuan Tang, Bowen Liu, Xuan Liu, Dingyan Shang · 2026-05-29
arXiv:2605. 28918v1 Announce Type: new Abstract: For sparse, structured reinforcement-learning tasks with semantic reward-function interfaces, LLM-generated reward shaping is better framed as debugging than one-shot generation.
Read next because When LLM Reward Design Fails: Diagnostic-Driven Refinement for Sparse Structured RL overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, under, eval, rate, control, without, trained, test. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28918v1 Announce Type: new Abstract: For sparse, structured reinforcement-learning tasks with semantic reward-function interfaces, LLM-generated reward shaping is better framed as debugging than one-shot generation. We study PPO-trained agents using MiniGrid as core evaluation and MuJoCo as boundary stress test. Our audit finds two dominant one-shot failure modes -- reward flooding and semantic/API misunderstanding -- plus a rarer weak-shaping case. We propose diagnostic-driven iterative refinement, where training diagnostics and a failure-mode taxonomy guide targeted reward-function revision. Refinement improves DoorKey-8x8 from 2.3% to 97.6% and KeyCorridor from 31.2% to 86.7% with high seed-to-seed variance. Controls show these gains are not from retrying or extra training: metrics-only re-prompting yields large drops, while a static-vocabulary control recovers much of the gap (87.6%; 70.7%), showing the taxonomy prompt is a major mechanism and dynamic labels provide only partially isolated incremental evidence. Budget-matched and Best-of-3 comparisons separate refinement from selection and training-time effects. Component-removal tests, sensitivity analyses, and an audit against author labels provide converging evidence for the debugging interpretation while revealing calibration limits. Continuous-control results show the boundary: success-based diagnostics can misfire in dense-reward locomotion, and return-trend feedback removes one false-positive mechanism without robust gains. The low-call protocol is a cost contrast with population-based reward search, not a benchmark comparison. In four crossed-variance-design environments, point estimates suggest larger gains when LLM reward-function variance dominates but bootstrap intervals are wide. The method is bounded to sparse structured tasks with reliable interfaces under PPO; fields like event_text may help, hurt, or be neutral.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, evaluation, benchmark.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28912unread
Cycle-Space Informed Detection of Autoencoded Blind False Data Injection Attacks on Power Systems
Xin Li, Chenhan Xiao, Jonathan Cohen, Aviad Elyashar, Yang Weng, Rami Puzis · 2026-05-29
arXiv:2605. 28912v1 Announce Type: new Abstract: The rapid growth of AI-driven data centers and large-scale energy storage systems is increasing the reliance of power system operation on real-time measurement data and automated decision-making.
Read next because Cycle-Space Informed Detection of Autoencoded Blind False Data Injection Attacks on Power Systems overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, under, line, rate, does. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28912v1 Announce Type: new Abstract: The rapid growth of AI-driven data centers and large-scale energy storage systems is increasing the reliance of power system operation on real-time measurement data and automated decision-making. However, many existing detection methods rely on statistical or data-driven analysis of measurements and can fail when attackers exploit the same data structure to craft stealthy perturbations. To illustrate this limitation, we demonstrate a blind False Data Injection Attack (FDIA) in which an Autoencoder learns the measurement manifold and generates perturbations aligned with the Jacobian null space, thereby allowing the attack to evade both residual-based baddata detectors and time-series anomaly detectors. To mitigate data-driven FDIAs which exploit the null space, we propose a topology-informed Cycle-Space Detector (CSD) that leverages the Cycle-Space of the network to impose structural constraints that enhance null space estimation. In addition, we prove that by using the Minimum Cycle Basis (MCB), the proposed CSD achieves the optimal generalization error for attack detection. By exploiting topology-derived cycle constraints rather than relying solely on numerical null space estimation, the proposed method does not require precise line parameters and improves the separation between normal and attacked measurements. Simulation results on IEEE 14-, 30-, 57-, and 118-bus systems demonstrate that the proposed method effectively detects data-driven FDIAs under realistic measurement noise.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28909unread
Sequential Physics-Constrained Neural Operator Forward Modeling for the $\textit{Norne}$ Reservoir System
Clement Etienam, Juntao Yang, Oleg Ovcharenko, Nick Luiken, Tsubasa Onishi, Nefeli Moridis, Issam Said · 2026-05-29
arXiv:2605. 28909v1 Announce Type: new Abstract: We develop a comprehensive mathematical and computational framework for sequential surrogate modeling of three-phase black-oil reservoir dynamics using neural operators, with particular emphasis on Fourier Neural Operators (FNO) and their physics-informed variant (PINO).
Read next because Sequential Physics-Constrained Neural Operator Forward Modeling for the $\textit{Norne}$ Reservoir System overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, under, full, trained, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28909v1 Announce Type: new Abstract: We develop a comprehensive mathematical and computational framework for sequential surrogate modeling of three-phase black-oil reservoir dynamics using neural operators, with particular emphasis on Fourier Neural Operators (FNO) and their physics-informed variant (PINO). The application focus is the Norne benchmark reservoir, defined on a heterogeneous $46\times112\times22$ grid ($N=113,344$ cells), with a production history spanning $T=30$ timesteps covering 3298 days. Our theoretical contributions are organized around four interlocking problems: (1) functional-analytic formulation in a product-Sobolev-space setting, including well-posedness of the implicit timestep map and sharp local Lipschitz estimates; (2) covariate shift quantification, proving that the Wasserstein-2 distance grows as $W_2 \leq \varepsilon(L^n-1)/(L-1)$, with exponential population-risk discrepancy for $L>1$; (3) physics-constrained spectral stability, showing PINO training with $\lambda_R \geq \lambda^*_R$ reduces the learned Jacobian spectral radius to $\rho_F + C\lambda_R^{-1/2}$, yielding uniform-in-time rollout error $|\delta_n| \leq \varepsilon/(1-\rho)$; and (4) $K$-step TBPTT gradient analysis, deriving geometric bias decay $O(\rho^K)$, optimal window $K^ = O(\log(T/\sigma^2))$, and Adam convergence $O(1/\sqrt{t}) + O(\rho^{K^*})$. Empirical validation confirms all theoretical predictions: autoregressive PINO surrogates sustain $R^2>0.99$ (oil), $R^2>0.90$ (gas), $R^2\approx 0.80$ (pressure), and monotonically improving $R^2$ (water) across the full 3298-day horizon, trained on eight NVIDIA B200 GPUs in under one hour. A 1000-member ensemble runs in under one minute on a single B200 GPU, giving a ${\sim}10^4\times$ wall-clock speedup over the OPM finite-volume simulator.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias, benchmark.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28889unread
Context Distillation as Latent Memory Management
Ziyang Zheng, Zeju Li, Xiangyu Wen, Jianyuan Zhong, Junhua Huang, Lei Chen, Mingxuan Yuan, Qiang Xu · 2026-05-29
arXiv:2605. 28889v1 Announce Type: new Abstract: Context distillation compresses contextual information into model parameters, yet existing methods often ignore how multiple distilled latent memories should be stored, retrieved, and safely activated in non-oracle settings.
Read next because Context Distillation as Latent Memory Management overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, eval, line, candidate, lora, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28889v1 Announce Type: new Abstract: Context distillation compresses contextual information into model parameters, yet existing methods often ignore how multiple distilled latent memories should be stored, retrieved, and safely activated in non-oracle settings. We formulate context distillation as a latent memory management problem. We distill each context into an independent LoRA adapter, forming a modular memory bank that enables explicit memory selection. Given a query, our framework retrieves candidate memories, routes the query to the most suitable adapter, and uses a Self-Gating mechanism to decide whether latent memory should be activated. To improve efficiency, we further introduce cache sharing to reduce management overhead during inference. Experiments show that our method substantially outperforms baselines with retrieval, while Self-Gating improves robustness by deactivate unnecessary latent memories.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses robustness.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28873unread
Pre-Registering the Detectable Effect: A Paired-MDE Budget for 4-bit Quantization Benchmarks, with a Pilot Audit
Zexin Zhuang, Yanhang Li, Zhichao Fan · 2026-05-29
arXiv:2605. 28873v1 Announce Type: new Abstract: This is a planning-method note with an unpaired pilot audit.
Read next because Pre-Registering the Detectable Effect: A Paired-MDE Budget for 4-bit Quantization Benchmarks, with a Pilot Audit overlaps with clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: class, alpha, line, rate, does, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28873v1 Announce Type: new Abstract: This is a planning-method note with an unpaired pilot audit. We adapt the classical paired-binary sample-size calculation (Miettinen, 1968) to quantization benchmarks, giving a conservative minimum detectable effect (MDE) bound $\delta^{*} \le (z_{1-\alpha/2}+z_{1-\beta})\sqrt{\rho_d/m}$ in the paired item count $m$ and the FP16-NF4 disagreement rate $\rho_d$. The bound turns "how reliable is my quantization claim?" into a one-line budget a benchmark designer can commit to before running. We illustrate the bound on four models and four benchmarks ($k=5$ splits of $n=100$), and add a parallel MMLU prompt-template study to put the bound's quantization-noise scale alongside the prompt-noise scale. Assuming $\rho_d=0.10$ (an unmeasured planning value), all observed NF4-FP16 deltas fall below the implied MDE, and most cross-split SDs lie within $\pm 1.5$ pp of the binomial reference $\sqrt{p(1-p)/n}$, so much of the variance reported as "benchmark unreliability" on $n=100$ subsamples is binomial sampling noise. The single borderline cell (OPT-WinoGrande, $|\Delta|=3.2$ pp) is below the implied MDE at $\rho_d=0.10$ but above it at $\rho_d=0.05$, illustrating the planning trade-off the bound makes explicit. On MMLU, prompt-template ranges of 2-10 pp meet or exceed the largest observed quantization delta (3.2 pp), so a quantization audit that does not first fix the prompt template absorbs template variance into its noise floor. We complement the bound with a five-line pre-registration template.
Potential threat/caveat for clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28870unread
Representation Alignment Rests on Linear Structure
Kiril Bangachev, Guy Bresler, Yury Polyanskiy · 2026-05-29
arXiv:2605. 28870v1 Announce Type: new Abstract: We investigate the Platonic Representation Hypothesis (PRH) through a tripartite statistical framework of representations: signal, bias, and noise.
Read next because Representation Alignment Rests on Linear Structure overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, strong, text, word, alignment, line, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28870v1 Announce Type: new Abstract: We investigate the Platonic Representation Hypothesis (PRH) through a tripartite statistical framework of representations: signal, bias, and noise. {1) Signal:} We propose that Platonic alignment arises from the universal relationship between objects and attributes, which is encoded linearly in representations according to the Linear Representation Hypothesis (LRH). We provide evidence that LRH helps explain PRH by extracting linear object-attribute features with sparse autoencoders and showing that these sparse representations often exhibit stronger cross-modal alignment than their dense counterparts. {2) Bias:} Models have different implicit biases due to the diverse architectures and training procedures used. We show that this difference can be partially mitigated. Centering and normalization consistently improve cross-model alignment. {3) Noise:} Finite-sample training leads to noise in representations. We provide evidence that representational noise is driven by data scarcity by revealing a strong and consistent positive correlation between word frequency and alignment in LLMs and text embedding models. Synthesizing signal, bias, and noise, we propose a statistical model that refines the Linear Representation Hypothesis and explains further phenomena related to the alignment of representations emerging from diverse modern AI architectures.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28867unread
PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation
Junru Zhang, Lang Feng, Jinbo Wang, Xu Guo, Yucheng Wang, Han Yu, Min Wu, Yabo Dong, Duanqing Xu · 2026-05-29
arXiv:2605. 28867v1 Announce Type: new Abstract: Generating high-quality time-series data is challenging because real-world signals often exhibit multimodal patterns and multiscale dynamics, including oscillations and high-frequency variations.
Read next because PrismFlow: Residual Dynamics for Flow Matching in Time-Series Generation overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: text, rect, correct, line, implement, trained, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28867v1 Announce Type: new Abstract: Generating high-quality time-series data is challenging because real-world signals often exhibit multimodal patterns and multiscale dynamics, including oscillations and high-frequency variations. Flow Matching (FM) offers an efficient alternative to diffusion models, but practical implementations typically rely on a single finite-capacity global vector-field estimator. In such heterogeneous temporal distributions, distinct regimes may pass through nearby flow states while requiring incompatible conditional velocities. A monolithic estimator trained with the standard $\ell_2$ velocity-matching objective may therefore learn an overly smoothed approximation of the local transport field. This estimator-level smoothing can attenuate branch-specific dynamics, leading to spectral distortion and poor mode coverage. To address this, we propose PrismFlow, a new FM method with Koopman-inspired dynamical experts. Each expert learns residual corrections in a latent space where local nonlinear temporal evolution can be approximated by linear transitions. We further propose a confidence-aware Winner-Take-All (WTA) objective that updates only the expert best aligned with each sample while masking gradients to the others, encouraging mode-specific specialization. During sampling, the selected expert adds a residual dynamical correction to the global transport field, preserving FM stability while recovering fine-grained and high-frequency temporal structures. Across various benchmarks, PrismFlow effectively mitigates the spectral contraction in standard FM and achieves state-of-the-art performance, with a 15.6% gain in Context-FID and a 38.6% improvement in Discriminative Score, while remaining robust in low-data settings and effective for forecasting and imputation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28866unread
Continuity and Ordinality Matter: Constraining Time Series Tokens for Effective Time Series Analysis with Large Language Models
Musheng Li, Ziying Zhang, Cheng jin, Yuantao Gu · 2026-05-29
arXiv:2605. 28866v1 Announce Type: new Abstract: Token-based time series large language models (TS-LLMs) have emerged as a promising direction for time series analysis and reasoning.
Read next because Continuity and Ordinality Matter: Constraining Time Series Tokens for Effective Time Series Analysis with Large Language Models overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, strong, rect, token, rate, stage, language, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28866v1 Announce Type: new Abstract: Token-based time series large language models (TS-LLMs) have emerged as a promising direction for time series analysis and reasoning. However, prior studies largely overlook the inherent continuity and ordinality of time series tokens, which substantially limits model performance. In this paper, we argue that preserving these properties in time series token embeddings is crucial for the effectiveness of token-based TS-LLMs. To this end, we propose COM (Continuity and Ordinality Matter), a continuity- and ordinality-aware strategy that integrates geometric constraints into both the initialization and training stages. Empirical results on multiple time series analysis benchmarks demonstrate that COM consistently improves the performance of token-based TS-LLMs, achieving competitive results and strong generalizability. Code is available at https://anonymous.4open.science/r/COM .
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28865unread
Emergent Semantic Representations in World Models through Physical Interaction without Linguistic Supervision
Jiayi Fang · 2026-05-29
arXiv:2605. 28865v1 Announce Type: new Abstract: What does a world model learn from physical exploration, without any linguistic supervision?
Read next because Emergent Semantic Representations in World Models through Physical Interaction without Linguistic Supervision overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: code, rect, alignment, without, does, position, lora, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28865v1 Announce Type: new Abstract: What does a world model learn from physical exploration, without any linguistic supervision? We argue the answer is organized by a single principle: the geometric structure of the physical world. Training a VAE-based world model on random embodied exploration, we find that its latent space develops spatial semantic structure that mirrors physical geometry -- direction accuracy 0.677+-0.029 versus 0.547 for a randomly initialized encoder, and position RSA 0.192+-0.047 versus 0.029 for random encoders (6.6x improvement), showing that training induces genuine structural organization beyond CNN inductive bias. Across 20 temporal checkpoints, prediction performance and semantic alignment co-improve (Spearman r=-0.61, p=0.004), consistent with the shared-driver account. We confirm this through a double knockout: standard KL regularization (beta=0.1) forces the encoder away from geometric structure, and both prediction performance and semantic alignment collapse simultaneously to near-chance by step 50,000 -- exactly as the shared-driver account predicts. Reducing beta to 0.001 restores geometric access and recovers both capabilities together. These findings establish physical world geometry as the organizing principle of world model representations, with direct implications for the design of semantically grounded embodied agents.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28863unread
Self-Play Reinforcement Learning under Imperfect Information in Big 2
Aalok Patwa · 2026-05-29
arXiv:2605. 28863v1 Announce Type: new Abstract: Imperfect-information multiplayer games test whether agents can act under hidden information, sparse rewards, and non-stationary opponents.
Read next because Self-Play Reinforcement Learning under Imperfect Information in Big 2 overlaps with clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)", clean result "Leakage rate is a usable signal for recovering trigger-shaped phrases on Gaperon-1125-1B without knowing the hidden trigger itself (MODERATE confidence)", clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: strong, under, eval, rate, control, test. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28863v1 Announce Type: new Abstract: Imperfect-information multiplayer games test whether agents can act under hidden information, sparse rewards, and non-stationary opponents. We study these challenges in Big 2, a four-player imperfect-information card game. We develop a self-play RL framework for Big 2 that enables controlled comparisons between policy-gradient and value-approximating agents. Under a common environment, input representation, training budget, and evaluation protocol, PPO outperforms Monte Carlo Q approximation, SARSA, and Q-learning against random, greedy, and heuristic Big 2 opponents. We further find that moderate entropy regularization improves PPO by preventing the policy from becoming overly deterministic, and that current-policy self-play provides a stronger finite-budget curriculum than checkpoint self-play or fixed-opponent training. Together, these results show that Big 2 is a useful controlled setting for studying deep RL under imperfect information, multiplayer interaction, delayed rewards, and variable action sets.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits at 23.5% when a co-trained partner learns <A>..., vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28862unread
Molecular Lead Optimization via Agentic Tool Planning
Lingxiao Li, Haobo Zhang, Ruohao Fan, Bin Chen, Jiayu Zhou · 2026-05-29
arXiv:2605. 28862v1 Announce Type: new Abstract: Drug discovery is a lengthy and resource-intensive process composed of multiple stages.
Read next because Molecular Lead Optimization via Agentic Tool Planning overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Training one persona to emit a [ZLT] marker without bystanders adopting it has a one-cell-wide LR x epochs window on Qwen2.5-7B-Instruct (LOW confidence)", clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)". Matching terms: under, source, line, compare, binding, length, stage, candidates. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28862v1 Announce Type: new Abstract: Drug discovery is a lengthy and resource-intensive process composed of multiple stages. Among these stages, lead optimization plays a critical role in transforming early hit compounds into viable drug candidates. This stage requires improving ADMET-related properties through subtle structural refinement while preserving key molecular substructures responsible for binding affinity to disease targets. Recent advances in artificial intelligence have shown promise in accelerating various aspects of drug discovery; however, most existing approaches to lead optimization rely on one-step molecular optimization, which fail to account for the long-term consequences of sequential design decisions. To address this limitation, we propose TRACE, a trajectory-aware, LLM-reasoning agent for molecular lead optimization that formulates tool selection as a sequential decision-making problem over action trajectories. Given a lead molecule and an optimization objective, TRACE makes trajectory-aware decisions over molecular optimization tools, enabling forward-looking refinement under structural constraints. Experiments on multiple ADMET optimization tasks show that our agent achieves higher optimization success, larger property improvements, and higher validity, while preserving molecular similarity compared to baseline models.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses limitation.
source ↗open
score 100arxiv cs.LG (Machine Learning)arxiv:2605.28850unread
Representation Signatures and Risk-Feedback Alignment in LLM Trading Agents
Weicheng Xue · 2026-05-29
arXiv:2605. 28850v1 Announce Type: new Abstract: We study behavioral alignment and representation dynamics of large language model (LLM) agents in financial decision environments.
Read next because Representation Signatures and Risk-Feedback Alignment in LLM Trading Agents overlaps with clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)", clean result "Coupling evil personas with wrong answers fails to protect Qwen2.5-7B from EM-induced alignment collapse — and the apparent capability ordering across coupling conditions is mostly eval contamination (LOW confidence)", clean result "Only continuous soft prefixes hit both EM axes at once on Qwen-2.5-7B-Instruct: discrete prompt searches split between the alignment objective and the distributional objective, and both discretizations of the soft prefix collapse (MODERATE confidence)". Matching terms: under, alignment, line, rate, control, without, does, position. Source: arxiv cs.LG (Machine Learning).
arXiv:2605.28850v1 Announce Type: new Abstract: We study behavioral alignment and representation dynamics of large language model (LLM) agents in financial decision environments. Using TradeArena, an auditable trading-agent testbed with risk reports, execution simulation, memory, and replayable trajectories, we analyze how rationales, positions, and interventions evolve under market stress. We find measurable pre-failure signatures: planning embeddings drift from normal-state centroids, fused plan-risk representations separate normal from pre-drawdown states, and manifold diagnostics show effective-rank contraction before failures. To address small-sample and embedding-choice concerns, we use 80 rolling failure anchors across eight LLM trajectories and show that contraction persists across hash, LSA, Transformer, and white-box hidden-state probes. Stress tests with CoT-free target weights, lexical controls, OHLCV noise, and false-audit reports indicate that rationale-level contraction can vanish without rationales, while intent-space contraction may remain; lexical diversity does not collapse; and fused signatures remain informative under noise. We also find that structured risk feedback can act as an external alignment signal without fine-tuning, but not as a universal performance enhancer: true audit feedback improves calibration for some models, return and drawdown for others, and reveals cases where hidden or placebo feedback has higher short-horizon return but weaker alignment diagnostics. Finally, a 51-stock intraday experiment reveals a correlation blind spot: LLM rationales often justify concentrated exposure to coupled assets that the risk layer repeatedly clips, with a rolling Markowitz baseline as a covariance reference. These results support a research claim rather than a profitability claim: auditable risk feedback and representation trajectories reveal when LLM financial reasoning is aligning, drifting, or failing.
Potential threat/caveat for clean result "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)": this item discusses failure, failures.
source ↗open
score 92arxiv stat.ML (Machine Learning)arxiv:2605.29464unread
Deep Optimal Individualized Treatment Rules for Bivariate Survival Outcomes via Adaptive Prediction-Powered Learning
Kun Ren, Yifan Cui, Wen Su · 2026-05-29
arXiv:2605. 29464v1 Announce Type: new Abstract: In randomized trials involving multiple treatments, bivariate survival outcomes present significant analytical challenges for making decisions.
Read next because Deep Optimal Individualized Treatment Rules for Bivariate Survival Outcomes via Adaptive Prediction-Powered Learning overlaps with clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)", experiment "#351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation", experiment "Language-mismatch LoRA SFT on Qwen2.5-7B leaks the trained completion language into bystander directives the model was never trained on, absent under same-language SFT (LOW confidence)". Matching terms: rate, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2605.29464v1 Announce Type: new Abstract: In randomized trials involving multiple treatments, bivariate survival outcomes present significant analytical challenges for making decisions. This paper addresses the problem of deriving optimal individualized treatment rules to maximize the joint survival probability beyond fixed time points $(t_1, t_2)$ through deep neural networks, while accounting for right censoring. We propose a novel approach that models treatment rules via stochastic policies, coupling marginal accelerated failure time models via link function to capture bivariate dependence. To enhance robustness and effectiveness of decision making, we introduce an adaptive prediction-powered method that leverages auxiliary predictions from machine learning models.
Potential threat/caveat for clean result "The marker is a representational handle, not a behavioural one — sharing it between a villain persona and the assistant transfers no misalignment (HIGH confidence)": this item discusses failure, robustness.
source ↗open

Methods

score 38M7 QA inline RSS threat sourceunread
Artifact verification caveats for Sagan clean results
M7 QA · No release date
This paper studies failure modes and caveats when Sagan creates a clean result only after verifying an artifact row. It proposes benchmark checks for artifact verification, clean-result review comments, and negative controls.
My work produces clean results with explicit confidence labels (MODERATE, LOW, HIGH) for findings like marker leakage rates and backdoor firing conditions — the verification caveats described here apply directly to how trustworthy those confidence assignments actually are.
This paper studies failure modes and caveats when Sagan creates a clean result only after verifying an artifact row. It proposes benchmark checks for artifact verification, clean-result review comments, and negative controls.
Potential threat/caveat for experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone": this item discusses failure, caveat, caveats, negative, benchmark.
source ↗open