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 cs.CR (Cryptography and Security)arxiv:2606.06875unread
Unified Safe In-context Image Generation in Multimodal Diffusion Transformers via Restricting Unsafe Information Flows
Xiang Yang, Feifei Li, Mi Zhang, Geng Hong, Xiaoyu You, Mi Wen, Min Yang · 2026-06-08
arXiv:2606. 06875v1 Announce Type: cross Abstract: Diffusion transformers (DiTs) equipped with multimodal attention (MM-Attn) have become a dominant paradigm for image generation.
Read next because Unified Safe In-context Image Generation in Multimodal Diffusion Transformers via Restricting Unsafe Information Flows overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, propagate, stage. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06875v1 Announce Type: cross Abstract: Diffusion transformers (DiTs) equipped with multimodal attention (MM-Attn) have become a dominant paradigm for image generation. However, preventing the generation of harmful content remains a critical challenge, particularly in image-to-image (I2I) editing tasks. Existing safety mechanisms are primarily designed for text-to-image (T2I) synthesis or U-Net-based architectures, which limits their effectiveness for unified safety mitigation in DiT-based frameworks. To bridge this gap, we propose Unified Visual Safety Regulator (UVR), a training-free safe generation framework that regulates unsafe semantics in generated images. UVR is grounded in an analysis of attention dynamics from the perspective of information flow in MM-Attn. We identify a task-independent start-up stage, during which unsafe semantics in output patches rapidly emerge and can be accurately localized, followed by task-specific semantic amplification and interference stages, where harmful signals are further propagated and entangled with benign content. Based on these observations, UVR mitigates unsafe generation through unified, targeted attention modulation and explicit restriction of harmful information flow over the identified unsafe output patches. Experiments across various concepts show that UVR achieves state-of-the-art safety performance by achieving 91% and 77% erase rate in image synthesis and editing tasks, while preserving visual quality and fidelity with minimal degradation. Code is available at https://github.com/deng12yx/UVR.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06700unread
The Economics of Proof-of-Useful-Work
Rafael Pass · 2026-06-08
arXiv:2606. 06700v1 Announce Type: cross Abstract: Proof-of-work (PoW) blockchains rely on computational expenditure to secure a ledger supporting a native cryptocurrency.
Read next because The Economics of Proof-of-Useful-Work 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, token, rate, without, does, chain, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06700v1 Announce Type: cross Abstract: Proof-of-work (PoW) blockchains rely on computational expenditure to secure a ledger supporting a native cryptocurrency. In existing systems such as Bitcoin, this expenditure is intentionally useless: the computation secures consensus but produces no external economic output. An emerging alternative -- proof of useful work (PoUW) -- enables the same computation to simultaneously secure the blockchain and generate economically valuable output. However, PoUW is often criticized on economic grounds: if the work is useful, attackers might be "paid to attack," potentially weakening security. We develop a competitive-equilibrium model of a PoUW blockchain in which compute can be allocated across pure mining, pure useful work -- instantiated as machine-learning inference -- or "duplex" work that produces both with computational overheads. We provide a complete closed-form characterization of equilibrium allocations and prices as a function of the duplex overheads and a single economic parameter -- the token-inference ratio -- measuring token adoption relative to the inference market. This characterization reveals three regimes: "Bitconia," in which the economy reduces to classical PoW; "Fortessia," in which duplex replaces mining, increasing security while useful output remains unchanged; and "Duplexia," in which token rewards subsidize inference, lowering prices and expanding inference supply. Contrary to the common strawman argument, PoUW does not make attacks economically cheap: once equilibrium prices are taken into account, the economic cost of a majority attack remains tied to the block reward. Moreover, in Duplexia, block rewards act as rebates on inference prices, generating additional socially useful computation that would not arise without the blockchain -- an expansion monotonically increasing in token adoption and technological efficiency.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06552unread
Beyond the Canonical Protocol: Quantum Encrypted Cloning from Secret-Sharing Access Structures
Gabriele Gianini, Stelvio Cimato, Jianyi Lin, Omar Hasan, Corrado Mio, Ernesto Damiani · 2026-06-08
arXiv:2606. 06552v1 Announce Type: cross Abstract: Quantum encrypted cloning shows that an unknown quantum state can be distributed into multiple encrypted copies without contradicting the no-cloning theorem: each copy is unusable on its own, but can be redeemed together with a suitable quantum key.
Read next because Beyond the Canonical Protocol: Quantum Encrypted Cloning from Secret-Sharing Access Structures overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, rate, extraction, without, does, language, never. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06552v1 Announce Type: cross Abstract: Quantum encrypted cloning shows that an unknown quantum state can be distributed into multiple encrypted copies without contradicting the no-cloning theorem: each copy is unusable on its own, but can be redeemed together with a suitable quantum key. Recent work has related canonical encrypted-cloning protocols to particular forms of quantum secret sharing. Here we take the converse perspective: instead of mapping a given encrypted-cloning protocol into QSS, we use QSS access structures as a design library from which encrypted-cloning schemes can be extracted. The criterion is access-structural. A QSS scheme supports a quantum encrypted-cloning structure whenever it contains a family of qualified sets with a non-qualified common intersection. The common subsystem is interpreted as the key, while the non-common parts are interpreted as encrypted clones relative to that key. Thus quantum encrypted cloning does not require a new notion of recoverability beyond QSS; what changes is the operational reading of QSS constituents as a mechanism for delayed and alternative redemption opportunities. This viewpoint separates redemption from perfect secrecy. Perfect QSS yields encrypted-cloning schemes with forbidden non-qualified subsystems, whereas ramp QSS naturally allows intermediate, partially informative non-redeeming subsystems. The resulting framework broadens quantum encrypted cloning from a specific protocol to a general access-structure primitive. We illustrate the extraction principle with threshold-like, ramp, hierarchical, and compartmented architectures, showing how encrypted clones may be symmetric or asymmetric, individual or composite, perfectly hidden or leaky. Equivalently, these constructions can be viewed as overlapping erasure-recovery regions of an isometric quantum code. This establishes secret sharing as a systematic design language for encrypted quantum redundancy.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.05273unread
Online Safety Regulation Increases Privacy Risk: Evidence from the UK Online Safety Act
Dhyey Mehta (University of Edinburgh), Eldar Jalilzade (Newcastle University), Maksim Kalameyets (Newcastle University), Rebecca Owens (Durham University), Marc Juarez (University of Edinburgh), Stergios Aidinlis (Durham University), Lei Shi (Newcastle University), Tu\u{g}rulcan Elmas (University of Edinburgh) · 2026-06-08
arXiv:2606. 05273v1 Announce Type: cross Abstract: Governments worldwide are increasingly regulating digital platforms to reduce online harms, particularly those affecting children.
Read next because Online Safety Regulation Increases Privacy Risk: Evidence from the UK Online Safety Act overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, latin, line, rate, does. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.05273v1 Announce Type: cross Abstract: Governments worldwide are increasingly regulating digital platforms to reduce online harms, particularly those affecting children. However, access restrictions can alter user behaviour and introduce new privacy and security risks. The UK Online Safety Act (OSA), passed in October 2023, illustrates this trend: it extends age-assurance and safety requirements to social media, search, and pornography services, and rolled out in phases. Ofcom's illegal content enforcement duties came into force in March 2025, and mandatory age verification for adult content took effect in July 2025. This phased rollout enables real-time observation of behavioural responses to regulation. To address this, we analyse Reddit discourse across VPN and UK Politics communities and conduct a privacy-policy risk analysis of 69 unique VPN services. We find that each of these three milestones produced significant stepwise increases in VPN-related discussion on Reddit: among UK-based users, posts and comments explicitly about VPN use in a regulatory or privacy context rose by +100%, +217%, and +415% respectively. UK Politics communities showed even larger effects, with OSA-related political discourse rising by +213%, +545%, and +464%, respectively, among UK-based users. UK VPN search interest on Google rose by +89% at the age-verification deadline. Users primarily framed this response around privacy, surveillance, and distrust of age-verification intermediaries rather than simple access-seeking. Demand increased across low, medium, and high-risk VPNs, but the proportional distribution remained broadly stable. These findings suggest that online safety regulation can create secondary privacy costs even when it does not disproportionately shift attention toward higher-risk providers.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07470unread
Verifiable and Confidential DNN Inference on Low-End Edge Devices
Mohamed Khalil Kiri (EURECOM), Ivan De Oliveira Nunes (University of Zurich), Aur\'elien Francillon (EURECOM), Norrathep Rattanavipanon (PSU Phuket) · 2026-06-08
arXiv:2606. 07470v1 Announce Type: new Abstract: Deploying deep neural network (DNN) inference on low-end edge devices raises two key challenges: protecting model confidentiality against a potentially compromised edge system and enabling verifiable inference without incurring prohibitive overhead.
Read next because Verifiable and Confidential DNN Inference on Low-End Edge Devices overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, source, without, trained, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07470v1 Announce Type: new Abstract: Deploying deep neural network (DNN) inference on low-end edge devices raises two key challenges: protecting model confidentiality against a potentially compromised edge system and enabling verifiable inference without incurring prohibitive overhead. Existing approaches either house partial models and inference software within trusted execution environments (TEEs), resulting in high cost and an application-dependent trusted computing base (TCB), or execute in untrusted environments, providing little security. In this work, we present VECODI, a framework for verifiable and confidential DNN inference on constrained edge devices. At its core, VECODI introduces SHANGRI-LA, a new execution abstraction on TrustZone-M TEEs that establishes a third runtime environment with privileges strictly between the Secure and Non-Secure Worlds. VECODI leverages SHANGRI-LA to execute untrusted inference code in the Non-Secure World while using minimal application-agnostic Secure-World support to protect model confidentiality and enable verifiability (with respect to proper execution of inference code and model parameters) of inference results. We realize VECODI on a real-world NUCLEO-L552ZE-Q development board and open-source its prototype. Our results show VECODI's small TCB, memory footprint, and runtime overhead, making it a practical option for secure inference in low-end edge devices.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07420unread
Lost in Migration: Exposing Android Framework Vulnerabilities in Parallel Java-Kotlin Implementations
Rui Li, Wenrui Diao, Debin Gao · 2026-06-08
arXiv:2606. 07420v1 Announce Type: new Abstract: Android has adopted Kotlin alongside Java across apps and core system components.
Read next because Lost in Migration: Exposing Android Framework Vulnerabilities in Parallel Java-Kotlin Implementations overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, source, rate, implement, compare, project. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07420v1 Announce Type: new Abstract: Android has adopted Kotlin alongside Java across apps and core system components. During this shift, we observe parallel implementations in the Android Open Source Project (AOSP) where the same component is implemented in both Java and Kotlin. In principle, their functional purposes are identical. In practice, subtle semantic divergences can appear. Such divergences are not vulnerabilities by themselves, but they provide useful clues that may reveal flaws in surrounding enforcement logic. To the best of our knowledge, this paper presents the first systematic study of Java-Kotlin parallel implementations in the Android framework and examines their security implications. We design and build ParaDroid, an analysis framework that identifies parallel methods at scale and compares their behaviors. ParaDroid normalizes code into a bytecode-level intermediate representation, reconstructs class-to-source mappings, and uses large language models to reason about method semantics and identify behavioral divergences. Evaluated on AOSP Android 14-16, ParaDroid identified 329 parallel method pairs and 37 vulnerable divergences. We responsibly disclosed the exploitable issues to the Android Security Team. Three vulnerabilities and two bugs have been confirmed, and two CVE IDs have been assigned. Our results demonstrate that parallel Java-Kotlin code paths provide a practical surface for discovering security flaws in modern Android.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07319unread
Authorized and Verifiable Searchable Encryption Based on Public Key Equality Test for Cloud Storage
Xiuping Li, Kaiwen Wang, Xiaolin Chang · 2026-06-08
arXiv:2606. 07319v1 Announce Type: new Abstract: Cloud storage revolutionizes data management but raises conflicts between functionality and privacy.
Read next because Authorized and Verifiable Searchable Encryption Based on Public Key Equality Test for Cloud Storage overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, token, rate, control, without, test. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07319v1 Announce Type: new Abstract: Cloud storage revolutionizes data management but raises conflicts between functionality and privacy. Public Key Encryption with Equality Test (PKEET), an advanced cryptographic technique, can enable multi-user searchable encryption (SE) through cross-key ciphertext comparison without shared keys. However, existing PKEET-based SE schemes lack ciphertext-file-level authorization, public verifiability, or SE-level support. This paper first proposes a novel PKEET scheme, AVPKEET (Authorized and Verifiable PKEET). It enables non-transferable and non-replayable authorization of ciphertext files, while supporting public verifiability, all without the need for trusted third parties. Then we propose an AVPKEET-based SE scheme, denoted as AVSE (Authorized and Verifiable SE), featuring one-time non-transferable tokens bound to users and nonces, batch operations, and fine-grained access control (ALL, PARTIAL, SINGLE). We prove OW-CCA2 security, token unforgeability, and verification soundness under standard assumptions. Experiment results demonstrate that AVSE achieves the most compact token size (168 bytes) while uniquely providing both ciphertext-file-level authorization and public verification, with acceptable overhead for cloud storage deployment.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07158unread
Synthetic APTs: the Collapse of TTP-Based Attribution
Francesco Balassone, V\'ictor Mayoral-Vilches, Mar\'ia Sanz-G\'omez, Paul Zabalegui-Landa, Stefan Rass, Davide Quarta, Daniel Sanchez-Prieto, Marina Oteiza-\'Alvarez, Almerindo Graziano, Lauren Min Kim, MinSeok Choi · 2026-06-08
arXiv:2606. 07158v1 Announce Type: new Abstract: Cyber Threat Intelligence CTI attribution relies on identifying the Tactics, Techniques, and Procedures TTPs that distinguish one threat actor from another.
Read next because Synthetic APTs: the Collapse of TTP-Based Attribution overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, control, full, position, another, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07158v1 Announce Type: new Abstract: Cyber Threat Intelligence CTI attribution relies on identifying the Tactics, Techniques, and Procedures TTPs that distinguish one threat actor from another. This approach presupposes that each adversary leaves a recognizable operational fingerprint. This work investigates whether AI driven adversary emulation challenges that presupposition. We deploy agents from our Cybersecurity SuperIntelligence CSI framework, configured as five Advanced Persistent Threat APT groups, APT28, APT29, APT41, APT44, and Lazarus Group, against AI driven Defender agents across two cyber ranges provided by CYBER RANGES, equipped with defensive software Wazuh, Velociraptor, Elasticsearch and active AI driven defenders: an enterprise network and a military infrastructure. Across 20 experiments using two defender models, a binary pattern emerges: all 10 Enterprise range experiments resulted in compromise 2 to 12 hosts per experiment, while all 10 Military range experiments were successfully defended or resulted in stalemates, regardless of APT profile or defender model. In 8 of 10 Enterprise experiments, attackers independently weaponized the defender's own Velociraptor endpoint management platform as a command and control channel, a convergent behavior not encoded in any threat intelligence profile. We argue that in the AI era, wherein agents can be deployed provided the right models are available and subject to the right scaffolding and agentic configuration, the entry barrier for operating like a nation state APT collapses: beyond nation states, individuals can now act like commonly identified threat actors, and with it, fundamentally undermine TTP based attribution.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07150unread
From Privacy to Workflow Integrity: Communication-Graph Metadata in Autonomous Agent Interoperability
Bijaya Dangol · 2026-06-08
arXiv:2606. 07150v1 Announce Type: new Abstract: Agent-interoperability protocols such as A2A and MCP standardize what agents say to one another, but assume address-based transport over HTTP(S).
Read next because From Privacy to Workflow Integrity: Communication-Graph Metadata in Autonomous Agent Interoperability 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, binding, alone, chain, candidate, capability, another. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07150v1 Announce Type: new Abstract: Agent-interoperability protocols such as A2A and MCP standardize what agents say to one another, but assume address-based transport over HTTP(S). Such transports protect message content, increasingly with end-to-end encryption. What they leave in the clear is the communication graph: which agent contacts which, when, and how often. In agent systems this graph is more consequential than a privacy framing suggests. Endpoints are often capability-labeled, workflows are structured and chained, and interactions are coupled to real actions, so an observer recovers more than past relationships. It can infer the pending workflow, the task being assembled and the action likely to follow. At machine speed, it can act on that inference before the workflow completes. The threat is therefore one of workflow integrity, not privacy alone: predictive leverage over autonomous action. We give a threat model for the agent communication graph; identify what makes agent metadata distinctively revealing (semanticity, prospectivity, actuation); define transport- and bootstrap-layer privacy properties and weigh candidate transports (SimpleX/SMP, Tor, mixnets) against them; and present an A2A case study in which a metadata-protecting binding is expressible but surfaces the protocol's identity assumptions. We test these on a generative model anchored to a real A2A capture. From passive metadata alone, with no payloads, a classifier recovers a task's class well above chance, from only the workflow's opening; applied together, the properties drive that recovery sharply back toward chance. Beyond what an observer can recover, we measure the leverage of acting on the leak: from a workflow's opening and under a fixed budget, an adversary choosing which workflows to act on realizes in this model most of a clairvoyant attacker's advantage over a metadata-blind one, and the same properties suppress it.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07005unread
The Sound of Malware: A Memory Forensics Approach for Android Malware Analysis via Audio Signals
Silvia Lucia Sanna, Massimo Palozzi, Leonardo Regano, Riccardo Lazzeretti, Giorgio Giacinto · 2026-06-08
arXiv:2606. 07005v1 Announce Type: new Abstract: Android malware analysis is currently facing increasing challenges in achieving robust classification and detecting stealth attacks.
Read next because The Sound of Malware: A Memory Forensics Approach for Android Malware Analysis via Audio Signals overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, rect, rate, control, without, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07005v1 Announce Type: new Abstract: Android malware analysis is currently facing increasing challenges in achieving robust classification and detecting stealth attacks. Modern threats employ advanced evasion strategies such as code obfuscation, dynamic loading, packing, and even steganographic manipulation of traditional static and dynamic features. These techniques reduce the effectiveness of signature-based systems and degrade the reliability of Machine Learning models that depend on explicit semantic indicators such as permissions, API calls, or control-flow structures. In this work, we propose \approachname, a memory forensics malware detection framework that shifts the analysis perspective from semantic program modeling to signal-based structural representation. Both static bytecode and early-execution memory snapshots are transformed into audio waveforms through direct binary-to-waveform mapping, preserving low-level structural patterns without requiring disassembly or feature engineering. The resulting signals are processed using handcrafted spectral descriptors, Convolutional Neural Networks, and transformer-based embeddings. Experiments on CICMalDroid2020 dataset and VirusTotal malware demonstrate that \approachname achieves up to 98.0\% accuracy, outperforming static sonification and competitive state-of-the-art approaches.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06895unread
Blockchain Infrastructure for Intelligent Cyber--Physical--Social Systems:Post-Quantum Security, Interoperability, and Trustworthy Data Economies in the Era of Embodied AI
Song Guo, Huawei Huang, Dongping Liu, Aoyu Zhang, Luyao Zhang · 2026-06-08
arXiv:2606. 06895v1 Announce Type: new Abstract: The deployment of embodied artificial intelligence via world-model-based robotics presents a transformative opportunity for blockchain infrastructure, establishing urgent demand for trustworthy data provenance, cross-organizational governance, and incentive-compatible sharing across decentralized ecosystems.
Read next because Blockchain Infrastructure for Intelligent Cyber--Physical--Social Systems:Post-Quantum Security, Interoperability, and Trustworthy Data Economies in the Era of Embodied AI 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)", 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: source, line, rate, implement, chain, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06895v1 Announce Type: new Abstract: The deployment of embodied artificial intelligence via world-model-based robotics presents a transformative opportunity for blockchain infrastructure, establishing urgent demand for trustworthy data provenance, cross-organizational governance, and incentive-compatible sharing across decentralized ecosystems. Simultaneously, quantum computing advances recognized by the 2025 Nobel Prize in Physics and the Turing Award threaten the cryptographic primitives securing these data economies, creating an interdependent imperative: long-lived verification for embodied AI depends on crypto-agile architectures capable of withstanding quantum adversaries. This tutorial examines blockchain as the coordination layer bridging this dual transition, from financial substrate to foundational Cyber-Physical-Social Systems infrastructure that simultaneously secures against quantum cryptanalysis and enables scalable, trustworthy data economies. The session opens with an immersive AWS Braket demonstration engaging participants with superconducting, trapped-ion, and neutral-atom hardware to assess cryptographic threat timelines and witness ECDSA-to-post-quantum signature transitions. Five integrated modules progress from embodied AI and world-model requirements through quantum hardware reality and evidence-based security migration, to scalable cross-shard architectures via BrokerChain protocols, trustworthy data economies implementing Croissant metadata standards and robotic learning provenance, and industry ecosystem integration for multi-modal cloud deployment. By bridging quantum hardware realities with embodied AI data requirements, this tutorial charts blockchain as unified infrastructure for next-generation decentralized intelligent environments, providing open-source frameworks and roadmaps for architecting quantum-resistant, interoperable, and data-trustworthy systems.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06860unread
On the Incentive Compatibility of Block Propagation in Bitcoin
Fumichika Maeda, Akira Sakurai, Taishi Nakai, Kazuyuki Shudo · 2026-06-08
arXiv:2606. 06860v1 Announce Type: new Abstract: Bitcoin is permissionless and does not rely on any central administrator, which gives it strong censorship resistance.
Read next because On the Incentive Compatibility of Block Propagation in Bitcoin overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, compare, does, propagate, full, chain. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06860v1 Announce Type: new Abstract: Bitcoin is permissionless and does not rely on any central administrator, which gives it strong censorship resistance. At the same time, it is important to incentivize miners to behave in ways that align with the interests of the system as a whole. This paper asks whether miners are individually incentivized to propagate blocks, one of the most fundamental processes in Bitcoin. Miners collectively maintain the blockchain by generating blocks and disseminating them across the network. If miners have an incentive not to propagate some blocks, this would indicate a fundamental flaw in Bitcoin's incentive design. Although prior work has studied how propagation delays affect forks and mining rewards, it has not fully characterized miners' incentives to improve block propagation under different tie-breaking rules. To address this gap, we derive analytical reward expressions for each tie-breaking rule based on a blockchain network model that captures the effect of forks on mining fairness. These expressions explicitly characterize how block propagation delays, hashrate distribution, and tie-breaking rules jointly determine mining rewards. We then use them to analyze miners' incentives to improve block propagation. Our results show, for example, that miners have no mining-reward incentive to relay blocks generated by other miners. By contrast, under the first-seen rule, every non-majority miner is incentivized to receive other miners' blocks more quickly and to propagate its own blocks more quickly. Finally, we compare tie-breaking rules and identify a trade-off between propagation incentives and mining fairness. In particular, the first-seen rule provides the strongest incentives to reduce propagation delays, but it also worsens mining fairness the most.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06767unread
The Custody Envelope Threshold: Authority-Scaled Admission of External Artifacts in Institutional Infrastructure
Amadeus Brandes · 2026-06-08
arXiv:2606. 06767v1 Announce Type: new Abstract: Modern infrastructure depends on externally maintained artifacts such as package-registry dependencies, CI/CD actions, container images, Terraform providers and modules, developer extensions, model artifacts, and AI tool servers.
Read next because The Custody Envelope Threshold: Authority-Scaled Admission of External Artifacts in Institutional Infrastructure overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, test, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06767v1 Announce Type: new Abstract: Modern infrastructure depends on externally maintained artifacts such as package-registry dependencies, CI/CD actions, container images, Terraform providers and modules, developer extensions, model artifacts, and AI tool servers. These artifacts are easy to fetch but difficult for institutions to admit, govern, and revoke. This paper proposes the Custody Envelope Threshold, an authority-scaled model of artifact admission. It argues that direct institutional admission is defensible only when object identity, ingress path, and revocation capacity are sufficiently closed relative to the execution authority delegated to the artifact. When this threshold is not met, institutions tend to proxy, policy-mediate, vendor-mediate, internalize, quarantine, or reject the artifact. The framework is operationalized as a four-condition ordinal instrument and connected to reference-monitor reasoning, least privilege, and transaction cost economics. It is applied to package dependencies, GitHub Actions, container images, Terraform providers and modules, developer extensions, and open model artifacts, with Model Context Protocol (MCP) servers treated as held-out evidence. The paper also specifies a validation design, deterministic prediction function, and OSF replication package for testing whether high-scrutiny institutions converge toward stronger custody closure for high-authority artifacts.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06697unread
AgileOS: A GPU Operating System Layer for Protected CUDA Services
Zhuoping Yang, Yiyu Shi, Alex Jones, Peipei Zhou · 2026-06-08
arXiv:2606. 06697v1 Announce Type: new Abstract: Modern GPU applications increasingly interact with storage systems, network devices, vendor libraries, and GPU-resident services rather than executing only isolated compute kernels.
Read next because AgileOS: A GPU Operating System Layer for Protected CUDA Services overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06697v1 Announce Type: new Abstract: Modern GPU applications increasingly interact with storage systems, network devices, vendor libraries, and GPU-resident services rather than executing only isolated compute kernels. This shift creates a need for operating-system-like protection around GPU services, where service metadata, device queues, memory-mapped I/O regions, and library-internal state should not be directly exposed to untrusted application kernels. However, today's CUDA programming model, by default, still gives each application direct ownership of its CUDA context, device pointers, runtime handles, module loading path, and kernel launches, leaving protected GPU services to build their own ad hoc interfaces and isolation mechanisms. This paper presents the initial design and prototype scope of AgileOS, a GPU operating-system layer for protected CUDA services. AgileOS virtualizes CUDA at the library boundary: applications link against client-side CUDA Runtime, Driver, and selected library shims, while a trusted runtime worker owns the real CUDA context and mediates supported operations. To protect service state and module interfaces, AgileOS also defines a GPU memory-management model that separates user allocations from protected module/MMIO ranges, using pointer validation and memory access guards via PTX injection. AgileOS is modularized and flexible, supporting a range of protected services and existing libraries such as cuFFT and PyTorch. The prototype includes client-side interceptors, worker-side CUDA handlers, virtualized CUDA object tables, protected AgileOS modules, a GPU memory manager that separates user allocations from protected module/MMIO ranges, selected trusted library adapters, and the PTX-level kernel memory guard.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06570unread
MalTree: Tracing Malware Evolution from Embeddings at Scale
Akash Amalan, Georgios Smaragdakis, Tom J. Viering · 2026-06-08
arXiv:2606. 06570v1 Announce Type: new Abstract: Malware detection remains largely reactive: machine learning models trained on known samples degrade as threats evolve.
Read next because MalTree: Tracing Malware Evolution from Embeddings at Scale 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, trained, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06570v1 Announce Type: new Abstract: Malware detection remains largely reactive: machine learning models trained on known samples degrade as threats evolve. Understanding evolutionary relationships among malware families can inform proactive defense, but traditional reverse engineering can take months to years to uncover such lineage relationships. We propose MalTree, a framework that applies bioinformatics inspired phylogenetic techniques (UPGMA and Neighbor-Joining) at scale to model malware evolution automatically using structural, behavioral, and image-based features. We introduce temporal validation using VirusTotal timestamps to assess whether inferred trees reflect actual evolutionary order. MalTree achieves 87% temporal consistency, indicating that inferred evolutionary relationships closely align with real-world emergence timelines. Our analysis shows that some families mutate over 10 times faster than others, suggesting that detection strategies should be tailored to family-specific evolutionary tempos. Case studies, including the Mirai botnet, confirm that inferred relationships from our phylogenetic tree align with documented threat intelligence. Our framework provides a foundation for shifting malware analysis from sample-by-sample classification toward lineage-aware evolutionary modeling.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07113unread
Beyond Post-hoc Explanation: Toward Glassbox AI via Probabilistic Mediation
Manuele Leonelli · 2026-06-08
arXiv:2606. 07113v1 Announce Type: new Abstract: Large language models are rapidly becoming infrastructural components in high-stakes institutional settings, including public administration, legal reasoning, and healthcare, where opacity is not merely inconvenient but institutionally and legally untenable.
Read next because Beyond Post-hoc Explanation: Toward Glassbox AI via Probabilistic Mediation overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, test, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07113v1 Announce Type: new Abstract: Large language models are rapidly becoming infrastructural components in high-stakes institutional settings, including public administration, legal reasoning, and healthcare, where opacity is not merely inconvenient but institutionally and legally untenable. Existing approaches to explainability are predominantly post-hoc, offering unstable, non-contestable accounts that have no formal relationship to the reasoning process that produced the output. We argue that the problem is not the absence of explanation but the absence of structured reasoning in the first place. This paper makes the case for a fundamentally different architecture, which we call the Glassbox Framework, in which Bayesian networks serve as transparent, ante-hoc mediation layers for generative models. Bayesian networks encode domain knowledge, causal assumptions, and probabilistic dependencies before inference occurs, enabling auditable reasoning traces, uncertainty quantification, and contestable outputs. We characterise the architecture of this framework and ground it in a benefit eligibility scenario, identifying the foundational challenges spanning semantic alignment, dynamic model construction, probabilistic grounding, and human governance that must be solved to realise it at scale. By shifting from post-hoc explanation to ante-hoc probabilistic mediation, this work outlines a principled path toward AI systems that are not only powerful but fundamentally accountable.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06893unread
Workflow-to-Skill: Skill Creation via Routing-Workflow-Semantics-Attachments Decomposition
Yuyang Zhang, Xinyuan Han, Xudong Jiang, Run Wang · 2026-06-08
arXiv:2606. 06893v1 Announce Type: new Abstract: Large language model agents increasingly rely on Skills to encode procedural knowledge, yet high-quality Skills remain costly to hand-write.
Read next because Workflow-to-Skill: Skill Creation via Routing-Workflow-Semantics-Attachments Decomposition overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, control, position, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06893v1 Announce Type: new Abstract: Large language model agents increasingly rely on Skills to encode procedural knowledge, yet high-quality Skills remain costly to hand-write. This paper studies automatic Skill construction from heterogeneous interaction evidence, including demonstrations, agent trajectories, tool traces, and execution logs. We argue that trace-to-skill construction is not simple summarization tasks, because traces are fragmented, redundant, and may miss rare but safety-critical behaviors. To address this, we introduce RWSA, a workflow-oriented intermediate representation that decomposes Skills into Workflow structure, execution Semantics, and runtime Attachments, capturing task decomposition, control flow, verification, safety, rollback, and state management. Building on RWSA, we propose W2S, a framework that segments traces, induces local Skill drafts, aligns shared structures, reconciles branches, and compresses redundancy while preserving evidence and confidence annotations. Experiments on 70 Skills show that W2S improves behavioral replay consistency by 10.5% over summarization- and prompting-based baselines, highlighting the need to treat traces as executable runtime specifications rather than compressible text.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06531unread
CARVE-Q: Quantum-Proposed, Classically Certified Interactive Driving Repair
Yifan Wang · 2026-06-08
arXiv:2606. 06531v1 Announce Type: new Abstract: The critical question after a correct driving veto is not only whether a maneuver is unsafe, but whether the blocked interaction admits a lawful, auditable, and responsibility-bounded repair.
Read next because CARVE-Q: Quantum-Proposed, Classically Certified Interactive Driving Repair 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, rate, binding, emit, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06531v1 Announce Type: new Abstract: The critical question after a correct driving veto is not only whether a maneuver is unsafe, but whether the blocked interaction admits a lawful, auditable, and responsibility-bounded repair. Prediction and game-theoretic planners can suggest plausible cooperation, yet they do not return a proof that the repair respects hard rules, right-of-way, cost allocation, and ego fallback. We introduce CARVE, Certified Affordable Repair of Vetoed maneuvers via Envelopes, a certificate architecture for prediction-free interactive repair. Given a vetoed maneuver, CARVE constructs a finite repair lattice and emits a structured certificate recording the binding rule, selected joint repair, right-of-way-scaled cooperation envelope, responsibility-weighted cost split, and ego-only fallback. This certificate view reveals the algorithmic bottleneck: multi-owner repair induces a product lattice $M = \prod_j |\mathcal{A}_j|$. We therefore introduce CARVE-Q, a verifier-shielded quantum-AI search layer that applies quantum minimum finding only to this black-box lattice while leaving all safety authority classical. In the conservative verifier-oracle model, exact classical minimum finding requires $\Theta(M)$ queries in the worst case, whereas Durr-Hoyer/Grover minimum finding uses $O(\sqrt{M})$ oracle queries with high probability. We prove verifier-shielded certificate soundness, priority non-elicitation, black-box query separation, and finite-precision reversible-oracle constructibility. We then demonstrate state-vector minimum finding on CARVE repair oracles up to 65,536 assignments and validate certificate preservation on Lanelet2-grounded INTERACTION replay with 100% right-of-way respect, 100% blame consistency, and zero priority false positives. The result is a trust-bounded quantum-AI pattern for certified autonomy: quantum proposes; CARVE certifies.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06519unread
SafeGene: Reusable Adapters for Transferable Safety Alignment
Yanghan Wang, Zhiqiang Kou, Fu Feng, Jing Wang, Xin Geng · 2026-06-08
arXiv:2606. 06519v1 Announce Type: new Abstract: Open-weight LLMs are increasingly fine-tuned into customized assistants, but downstream fine-tuning can weaken safety alignment and make models more vulnerable to malicious prompts, even when the training data is not intentionally harmful.
Read next because SafeGene: Reusable Adapters for Transferable Safety Alignment 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: alignment, assistant, rate, capability, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06519v1 Announce Type: new Abstract: Open-weight LLMs are increasingly fine-tuned into customized assistants, but downstream fine-tuning can weaken safety alignment and make models more vulnerable to malicious prompts, even when the training data is not intentionally harmful. This creates a recurring safety recovery problem as target models are repeatedly updated with new task data or user interactions. We propose SafeGene, a reusable safety-adapter module designed for cross-task reuse within each architecture-compatible model family. Rather than treating safety recovery as a model-specific repair step, SafeGene treats safety capability as an independent, reusable adapter representation decoupled from task-specific updates. This representation is obtained from aligned--degraded model discrepancies, refined into task-transferable safety vectors through data-aware layer selection, and expressed in each downstream task-adapted model via few-shot layer-wise coefficient recalibration. Experiments across multiple model families, downstream tasks, and safety judges show that SafeGene-enhanced models reduce harmful response rates while maintaining downstream performance, outperforming representative safe adaptation methods in safety--utility trade-off.
- score 100arxiv cs.CL (NLP)arxiv:2606.06906unread
EASE-TTT: Evidence-Aligned Selective Test-Time Training for Long-Context Question Answering
Xiaopeng Yuan, Zebin Wang, Suwen Wang, Zongxin Yang, Haohan Wang, Yushun Dong · 2026-06-08
arXiv:2606. 06906v1 Announce Type: new Abstract: Long-context question answering (QA) remains challenging for smaller language models even when answer-bearing evidence is already present in the input.
Read next because EASE-TTT: Evidence-Aligned Selective Test-Time Training for Long-Context Question Answering overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, soft, eval, token, line, control. Source: arxiv cs.CL (NLP).
arXiv:2606.06906v1 Announce Type: new Abstract: Long-context question answering (QA) remains challenging for smaller language models even when answer-bearing evidence is already present in the input. Existing within-context retrieval methods localize and expose candidate evidence chunks for the question, but they stop at input-level evidence exposure rather than adapting the query-side attention parameters that control how the model allocates attention over full-context positions. In contrast, lightweight test-time adaptation methods, such as query-only test-time training (qTTT), leave evidence localization unresolved because their generic span-level self-supervised objectives do not identify which context positions support the current answer. In this paper, we propose Evidence-Aligned SElective Test-Time Training (EASE-TTT), a within-context retrieval-augmented test-time training framework that converts selected evidence chunks into a soft attention supervision target over their token positions. Instead of replacing the full context with retrieved chunks, EASE-TTT uses the resulting attention target to guide query-side adaptation, with the adapted model generating the final answer from the original full context. Experiments on six LongBench QA tasks and three small decoder-only language models show that EASE-TTT achieves the strongest macro-average performance among full-context inference, retrieval-only baselines, and qTTT, supporting evidence-aligned test-time adaptation in long-context QA.
- score 100arxiv cs.CL (NLP)arxiv:2606.06857unread
Interpreting Brain Responses to Language with Sparse Features from Language Models
Michael A. Lepori, Kendrick Kay, Greta Tuckute · 2026-06-08
arXiv:2606. 06857v1 Announce Type: new Abstract: A central goal of cognitive neuroscience is to characterize the features that are represented by human language cortex.
Read next because Interpreting Brain Responses to Language with Sparse Features from Language Models overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, latin, alignment, alone, another, test, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06857v1 Announce Type: new Abstract: A central goal of cognitive neuroscience is to characterize the features that are represented by human language cortex. Artificial language models (LMs) have emerged as a powerful tool to address this challenge, but studies relating biological and artificial representations are often criticized as relating one black box to another. The present work introduces Augmented Sparse Encoding Models, an encoding framework that replaces dense LM hidden states with hierarchically-organized sparse autoencoder (SAE) features, while explicitly including surprisal as a predictor. Using this approach, we (i) produce interpretations of neural responses and (ii) test whether model-brain alignment reflects primary or idiosyncratic variation in LM representations. Using a high-field 7T fMRI dataset of eight participants listening to 200 linguistically diverse sentences, we first validate our modeling framework by recovering previous interpretations of voxel populations tuned to processing difficulty and meaning abstractness. We then interpret a previously-uncharacterized (but reliable) voxel population and find that it is tuned to people-related content. Next, we show that the fronto-temporal human language network is predicted by a common set of features across its constituent regions, but find that frontal regions are relatively well-explained by surprisal alone, even in the absence of LM-based features. Finally, we show that brain responses during language processing are not merely predictable from an arbitrary set of LM features. Rather, brain responses are best explained by the features that tend to capture the most general information encoded in LM representations, suggesting a nontrivial correspondence between brain and LM language representation.
- score 100arxiv cs.CL (NLP)arxiv:2606.06840unread
Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces
Debjyoti Saha Roy, Byron C. Wallace, Javed A. Aslam · 2026-06-08
arXiv:2606. 06840v1 Announce Type: new Abstract: Modern reasoning models offer surprisingly strong zero-shot performance on challenging multi-label tasks that require selecting a small set of relevant options from hundreds of thousands to millions of candidate labels.
Read next because Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, candidates, candidate, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06840v1 Announce Type: new Abstract: Modern reasoning models offer surprisingly strong zero-shot performance on challenging multi-label tasks that require selecting a small set of relevant options from hundreds of thousands to millions of candidate labels. We investigate how they achieve this mechanistically. We characterize reasoning as a two-phase process: A broad "shortlisting" of candidates followed by fine-grained reasoning over the resulting set. We provide evidence across a range of datasets that these steps can be isolated and are complementary. Using this characterization, we develop a mechanistic distillation strategy that consistently outperforms standard distillation.
- score 100arxiv cs.CL (NLP)arxiv:2606.06835unread
Translate-R1: Cost-Aware Translation Tool Use via Reinforcement Learning
Pratik Jayarao, Chaitanya Dwivedi, Himanshu Gupta, Neeraj Varshney, Adithya M Devraj, Meet Vadera, Priyanka Nigam, Bing Yin · 2026-06-08
arXiv:2606. 06835v1 Announce Type: new Abstract: The performance gap across languages in LLMs is well documented, and closing it natively requires pretraining or fine-tuning on corpora that, for most languages, do not exist.
Read next because Translate-R1: Cost-Aware Translation Tool Use via Reinforcement 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 "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, alone, full, trained, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06835v1 Announce Type: new Abstract: The performance gap across languages in LLMs is well documented, and closing it natively requires pretraining or fine-tuning on corpora that, for most languages, do not exist. Translation offers an alternative: converting an input into the model's dominant language unlocks its full capabilities at once. Applying translation to every input, however, is wasteful for languages the model already handles, while leaving the choice to the model fails in the opposite way, as LLMs are overconfident and skip the tool even when they cannot understand the input. Prior work resolves this with language-specific rules, domain heuristics, language identifiers, or external routers, each requiring manual engineering. We instead learn a single policy that decides when to translate from reward alone, developing language- and domain-adaptive introspection that assesses its own comprehension and invokes translation only when it cannot solve a task natively. Using data built by our answer-preserving translation pipeline, we continue RL on the post-trained Qwen3-4B across 22 languages in 3 resource tiers (High, Low, XLow) and 5 domains, and introduce confidence-gated GSPO for cost-sensitive tool use. The gated policy lifts reward over the baseline by +4.6 on High, +23.5 on Low, and +17.5 on XLow. Against an unconstrained policy that almost always translates, it preserves full reward at 63% of the cost and is Pareto-optimal across 87% of the cost-sensitivity range. Additionally, to simulate behavior on a completely unseen language, we create 2 synthetic languages, where our gated policy improves +18.7 over the overconfident baseline that underutilizes the tool even on these incomprehensible inputs. The policy transfers zero-shot to 9 held-out languages, and we analyze how tool use emerges over training, per language and per domain.
- score 100arxiv cs.CL (NLP)arxiv:2606.06834unread
The Dark Regulome: Disentangling Predictability from Regulation in Genomic Foundation Models
Chahat Baranwal, Aadtya Baranwal, Lakshya Nitin Tandon · 2026-06-08
arXiv:2606. 06834v1 Announce Type: new Abstract: High-grade gliomas integrate into neural circuits through functional synapses with neurons, raising the question of which noncoding elements shape synaptogenic gene expression in tumor cells.
Read next because The Dark Regulome: Disentangling Predictability from Regulation in Genomic Foundation Models overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, rate, control, alone, does. Source: arxiv cs.CL (NLP).
arXiv:2606.06834v1 Announce Type: new Abstract: High-grade gliomas integrate into neural circuits through functional synapses with neurons, raising the question of which noncoding elements shape synaptogenic gene expression in tumor cells. The regulatory program written across the dark genome, what we call the $\textit{dark regulome}$, is the natural substrate to probe, and sequence foundation models offer a zero-shot route through in-silico mutagenesis (ISM); yet likelihood-based scoring is tautologically coupled to local sequence predictability, leaving the regulatory interpretation underdetermined. Across three architecturally distinct foundation models (Caduceus-Ph, HyenaDNA, Enformer) and 30,448 dark genome elements at 92 glioma-relevant loci, we introduce a residualization-and-permutation diagnostic that separates predictability-driven from regulation-driven RIS variance. A sharp 10kb proximal-regulatory horizon survives every control we apply, but the LM-derived element-class hierarchy does not: a six-feature linear baseline matches Caduceus top-decile membership at AUC $= 0.985$. Cross-architecture decomposition cleanly separates a sequence-predictability layer (the two language models co-rank long well-predicted transposable elements) from a regulatory-output layer (Enformer alone retains residual cCRE-discriminative signal), with literally zero overlap between the two top-100 lists. Conservation, brain cis-eQTL, and STRING-PPI cross-checks then anchor what biology survives: top-100 elements across all three models are $3.3\times$ enriched per model for matching brain eQTLs ($p_\mathrm{emp} < 5\times 10^{-3}$), while a tempting transposable-element regulatory layer and a striking NRXN1+NLGN1 protein-pair convergence both fail proper permutation tests once those tests are constructed. We deliver the diagnostic as a general methodological tool for any ISM-based regulatory study.
- score 100arxiv cs.CL (NLP)arxiv:2606.06812unread
Quantifying Media Representation Dynamics Across 25 Years of News Reporting on Policing-related Deaths
Farhan Samir, Jappun Dhillon, Meghna Ravikumar, Syed Ishtiaque Ahmed, Vered Shwartz · 2026-06-08
arXiv:2606. 06812v1 Announce Type: new Abstract: We perform the largest known computational analysis of Canadian news narratives about police-involved deaths, spanning 4,000 articles from the last quarter-century.
Read next because Quantifying Media Representation Dynamics Across 25 Years of News Reporting on Policing-related Deaths overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06812v1 Announce Type: new Abstract: We perform the largest known computational analysis of Canadian news narratives about police-involved deaths, spanning 4,000 articles from the last quarter-century. We develop a novel computational model, PerspectiveGap, grounded in prior sociological work on media representation of policing. We find that reporting on police-involved deaths on average features perspectives from state bureaucrats at a rate nearly three times as much as perspectives from other members of the public, including relatives, community members, eyewitnesses, lawyers representing the family, or civil liberties groups. A considerable fraction of articles contain no points of view from civilian actors, though civilian representation has increased in recent years. Qualitatively, we find that state bureaucrats' accounts of these deaths tend to be clinical and procedural, while civilian discourse carries considerably more emotional valence. The PerspectiveGap framework developed here can be contextualized to other jurisdictions, offering a scalable approach for analyzing how media systems construct narratives around policing and accountability.
- score 100arxiv cs.CL (NLP)arxiv:2606.06794unread
TA-RAG: Tone-Aware Retrieval-Augmented Generation for Peer-Support Health Communication
Yong-Bin Kang, Anthony McCosker · 2026-06-08
arXiv:2606. 06794v1 Announce Type: new Abstract: Retrieval-augmented generation (RAG) successfully grounds large language model (LLM) outputs in trusted documents, but factual grounding alone is insufficient for sensitive peer-support health communication.
Read next because TA-RAG: Tone-Aware Retrieval-Augmented Generation for Peer-Support Health Communication overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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 "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: recipient, rect, eval, line, control, without, alone, full. Source: arxiv cs.CL (NLP).
arXiv:2606.06794v1 Announce Type: new Abstract: Retrieval-augmented generation (RAG) successfully grounds large language model (LLM) outputs in trusted documents, but factual grounding alone is insufficient for sensitive peer-support health communication. In domains such as HIV peer support, responses must also be accessible, stigma-free, empathetic, and tailored to the recipient. This paper presents TA-RAG, a lightweight, prompt-based tone-aware RAG framework that embeds explicit tone control into a RAG pipeline without requiring model fine-tuning. We operationalise tone across four core components: stigma-free rewriting, readability adjustment, recipient adaptation, and empathy rephrasing. We evaluate TA-RAG through component-level tests using questions derived from HIV Online Learning Australia (HOLA), UNAIDS terminology guidance, readability metrics, peer-support standards from National Association of People with HIV Australia (NAPWHA), and a public empathy dataset. Results show that the TA-RAG's components improve their targeted communication quality while preserving key content. These findings emphasise that prompt-based tone control is a potential direction for making RAG outputs suitable for sensitive peer-support health communication.
- score 100arxiv cs.CL (NLP)arxiv:2606.06781unread
When Better Codebooks Are Not Enough: Predictive Performance and Behavioral Reliability in LLM Political Event Coding
Zixian He, Bharath Raahul Murugesan, Patrick Brandt, Yibo Hu · 2026-06-08
arXiv:2606. 06781v1 Announce Type: new Abstract: High accuracy does not necessarily make an LLM a faithful coder.
Read next because When Better Codebooks Are Not Enough: Predictive Performance and Behavioral Reliability in LLM Political Event Coding overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, under, eval, source, control, does. Source: arxiv cs.CL (NLP).
arXiv:2606.06781v1 Announce Type: new Abstract: High accuracy does not necessarily make an LLM a faithful coder. This issue matters because many social-science studies rely on expert-written codebooks to turn text into structured data. We study this problem in political event coding, a challenging source-target relation classification task beyond ordinary sentence-level classification, where models must determine what one actor did to another using detailed coding rules. We test whether expert codebooks become more effective when operationalized into LLM-friendly forms with clearer definitions, examples, retrieved context, and rules for difficult cases. We then evaluate behavioral reliability under controlled changes to label names, codebook order, and label-definition mappings. Clearer codebooks substantially improve classification performance, especially for fine-grained event classification. However, these predictive gains do not fully translate into behavioral reliability. Models may produce valid labels and recover definitions while still failing behavioral reliability tests under controlled codebook changes. These findings suggest that codebook-guided LLM systems should be evaluated not only by accuracy, but also by whether they preserve the coding logic that makes coded outputs meaningful for social-science research.
- score 100arxiv cs.CL (NLP)arxiv:2606.06748unread
Evidence Graph Consistency in Retrieval-Augmented Generation: A Model-Dependent Analysis of Hallucination Detection
Jianru Shen · 2026-06-08
arXiv:2606. 06748v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) reduces but does not eliminate hallucination in large language models.
Read next because Evidence Graph Consistency in Retrieval-Augmented Generation: A Model-Dependent Analysis of Hallucination Detection 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, rate, does, full, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06748v1 Announce Type: new Abstract: Retrieval-Augmented Generation (RAG) reduces but does not eliminate hallucination in large language models. Existing detection methods rely on flat similarity between generated answers and retrieved passages, ignoring structural relationships among evidence pieces and answer claims. We propose Evidence Graph Consistency (EGC), a framework that constructs a local evidence graph per response and computes five structural consistency measures as hallucination indicators. Evaluated on the full question answering split of RAGTruth across six LLMs (5,767 responses), EGC reveals a consistent model-family split: graph consistency features show the expected diagnostic direction for hallucinations in Llama-2 models but exhibit systematic reversal in GPT-4, GPT-3.5, and Mistral-7B. This reversal suggests qualitatively different hallucination patterns across model families and indicates that embedding-based graph consistency cannot serve as a model-independent hallucination detection signal.
- score 100arxiv cs.CL (NLP)arxiv:2606.06745unread
When to Think Deeply: Inhibitory Deliberation for LLM Reasoning
Zhixuan He, Yue Feng · 2026-06-08
arXiv:2606. 06745v1 Announce Type: new Abstract: Reasoning Large Language Models can improve problem-solving performance through deliberative inference, but invoking slow reasoning for every input is computationally expensive and often unnecessary.
Read next because When to Think Deeply: Inhibitory Deliberation for LLM Reasoning overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, control, test. Source: arxiv cs.CL (NLP).
arXiv:2606.06745v1 Announce Type: new Abstract: Reasoning Large Language Models can improve problem-solving performance through deliberative inference, but invoking slow reasoning for every input is computationally expensive and often unnecessary. We propose IDPR, a framework for response-conditioned inhibitory deliberation. IDPR first generates a concise intuitive answer and then uses an inhibition controller to decide whether that specific response should be released or suppressed in favor of slow reasoning. Unlike input-only routers, the inhibition controller conditions on the fast answer and fast-side evidence, including confidence, logit margin, parseability, and generation cost. We train the controller from paired fast-slow outcomes and select the inhibition threshold on a held-out validation set under an accuracy-first slow-call budget. On a held-out 5,000-example mathematical reasoning test set, IDPR invokes slow reasoning on only 8.20% of examples and improves accuracy from 47.90% to 48.92%. Under the same slow-call budget, random routing decreases accuracy to 46.76%, while the strongest confidence-based baseline reaches 48.22%. IDPR also achieves the highest corrective precision, showing that response-conditioned inhibition better identifies fast answers that benefit from slow reasoning.
- score 100arxiv cs.CL (NLP)arxiv:2606.06712unread
Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation
Xingyu Su, Jacob Helwig, Shubham Parashar, Atharv Chagi, Lakshmi Jotsna, Degui Zhi, James Caverlee, Dileep Kalathil, Shuiwang Ji · 2026-06-08
arXiv:2606. 06712v1 Announce Type: new Abstract: We study the transformation of autoregressive models (ARLMs) into diffusion language models (DLMs).
Read next because Data-Efficient Autoregressive-to-Diffusion Language Models via On-Policy Distillation overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, token, rate, trained, on-policy, position, language. Source: arxiv cs.CL (NLP).
arXiv:2606.06712v1 Announce Type: new Abstract: We study the transformation of autoregressive models (ARLMs) into diffusion language models (DLMs). Rather than pretraining from scratch, prior work replaces the causal attention in ARLMs with bidirectional attention and then trains the resulting model using a DLM objective. However, these approaches incur two distribution shifts. First, transitioning from a next-token prediction objective to a DLM objective can discard knowledge acquired by the ARLM during training. Second, standard DLMs suffer from a train-inference mismatch, as the training loss is defined on randomly masked sequences rather than the trajectories encountered at inference produced by confidence-based decoding. To address both challenges, we introduce an On-Policy Diffusion Language Model (OPDLM) in which On-Policy Distillation (OPD) is employed for ARLM-to-DLM transformation. Specifically, OPDLM is trained via self-OPD, where the student, an ARLM with bidirectional attention, generates its own trajectories, and the teacher, the original frozen ARLM, distills its knowledge by providing target logits on these trajectories. By training directly in an on-policy manner, OPDLM eliminates the train-inference mismatch in DLMs, while distillation from the original model enhances knowledge retention from the ARLM. Empirical results demonstrate that OPDLM requires 15x to 7,000x fewer training tokens with strong performance across a wide variety of tasks. OPDLM avoids the prohibitive cost of DLM pretraining and positions DLM transformation as a form of ARLM post-training.
- score 100arxiv cs.CL (NLP)arxiv:2606.06667unread
The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment
Jiachen Zhao, Zhengxuan Wu, Aryaman Arora, Yiyou Sun, David Bau, Weiyan Shi · 2026-06-08
arXiv:2606. 06667v1 Announce Type: new Abstract: The mechanisms behind LLMs' broad over-generalization beyond training examples remain unclear.
Read next because The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment 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, prefix, token, without, trained, test, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06667v1 Announce Type: new Abstract: The mechanisms behind LLMs' broad over-generalization beyond training examples remain unclear. Emergent misalignment (EM) offers a striking case study: finetuning on narrow tasks induces broad misalignment to semantically-unrelated test domains. In this work, we propose the Piggyback Hypothesis: the chat-template tokens can piggyback the finetuned behaviour onto out-of-domain queries. We validate this hypothesis by showing that subtle perturbations to the prefix (tokens preceding all user queries), or patching the prefix representations with those from the unfinetuned model, can restore alignment without changing the user query. Building on this finding, we propose Token-Regularized Finetuning (TReFT), which regularizes specific token representations during training to mitigate EM. Across different models and multiple EM-inducing datasets, TReFT reduces EM while preserving in-domain learning. On Llama-3.1-8B finetuned on the legal domain, TReFT achieves 33.5% more EM reduction than data interleaving with a retain set of aligned examples. We further show that TReFT extends to other narrow-finetuning settings, including abstention, tool use, and refusal (off-topic generalization is reduced by 54.3% on average), supporting the Piggyback Hypothesis. Broadly, our work highlights that LLMs may learn and generalize in unintended ways and suggests a path toward more constrained finetuning. It also calls for further study of how shared input features can piggyback model behavior across domains.
- score 100arxiv cs.CL (NLP)arxiv:2606.06586unread
Improving Cross-Lingual Factual Recall via Consistency-Driven Reinforcement Learning
Jonathan von Rad, Louis Arts, George Burgess, Eleftheria Kolokytha, Harry O'Donnell, Ektor Oikonomidis Doumpas, Eduardo Sanchez, Yao Lu, Pontus Stenetorp · 2026-06-08
arXiv:2606. 06586v1 Announce Type: new Abstract: Large language models (LLMs) trained predominantly on English data encode substantial world knowledge, yet often fail to express it reliably in other languages, a phenomenon known as cross-lingual factual inconsistency.
Read next because Improving Cross-Lingual Factual Recall via Consistency-Driven Reinforcement Learning overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, compare, trained, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06586v1 Announce Type: new Abstract: Large language models (LLMs) trained predominantly on English data encode substantial world knowledge, yet often fail to express it reliably in other languages, a phenomenon known as cross-lingual factual inconsistency. To study and address this, we introduce PolyFact, a large-scale parallel multilingual factual QA dataset containing 100K Wikidata-grounded facts across 12 typologically diverse languages. Using PolyFact, we compare light continual pretraining (CPT), supervised fine-tuning (SFT), and reinforcement learning via Group Relative Policy Optimization (GRPO) for improving cross-lingual factual recall in Qwen-2.5-7B and OLMo-2-1124-7B. We find that GRPO consistently outperforms SFT, improving both cross-lingual consistency and generalization to unseen languages, while CPT on parallel data yields limited additional gains. Mechanistic analyses further show that GRPO reorganizes multilingual routing by reducing language specialization in MLP layers and attention heads, thereby promoting more shared cross-lingual representations. We release our code, models, and dataset.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06823unread
PandaAI: A Practical Agent CQ2 for Neuro-symbolic Data Analysis And Integrated Decision-Making in Quantitative Finance
Yuqi Li, Siyuan Liu, Bingjun Liu · 2026-06-08
arXiv:2606. 06823v1 Announce Type: new Abstract: While deep learning has excelled in various domains, its application to sequential decision-making in finance remains challenging due to the low Signal-to-Noise Ratio (SNR) and non-stationarity of financial data.
Read next because PandaAI: A Practical Agent CQ2 for Neuro-symbolic Data Analysis And Integrated Decision-Making in Quantitative Finance overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, alpha, rate, trained, capability, language, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06823v1 Announce Type: new Abstract: While deep learning has excelled in various domains, its application to sequential decision-making in finance remains challenging due to the low Signal-to-Noise Ratio (SNR) and non-stationarity of financial data. Leveraging the reasoning capabilities of Large Language Models (LLMs), we propose \textbf{PandaAI}, a closed-loop neuro-symbolic LLM agent with market regime modeling and constrained alpha generation, which bridges general LLM reasoning with financial rigor and suppresses the financial toxicity of LLM-generated outputs. To bridge the gap between general linguistic capability and financial rigor, we fine-tune a domain-specific LLM. Furthermore, we integrate this LLM into a modular architecture and form a closed-loop system. Unlike traditional models that optimize isolated prediction metrics, \textbf{PandaAI} is designed as a neuro-symbolic agent that navigates the complex, real-world financial environment with explicit risk awareness. Extensive experiments on CSI 300 stock data show that \textbf{PandaAI} achieves a $18.2\%$ higher Rank IC and $25.7\%$ lower maximum drawdown than state-of-the-art time-series models. Our constrained LLM generation and dual-channel adaptation method provide a general paradigm for LLM deployment in high-stakes sequential decision-making scenarios.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06820unread
SCALE: Scalable Cross-Attention Learning with Extrapolation for Agentic Workflow Scheduling
Zhifei Xu, Jierui Lan, Zixuan Liang, Aiji Liang, Jinxi He · 2026-06-08
arXiv:2606. 06820v1 Announce Type: new Abstract: Agentic Large Language Model (LLM) systems decompose complex tasks into workflow Directed Acyclic Graphs (DAGs) whose primitives must be scheduled on heterogeneous clusters.
Read next because SCALE: Scalable Cross-Attention Learning with Extrapolation for Agentic Workflow Scheduling 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, good, without, alone, does, trained, test. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06820v1 Announce Type: new Abstract: Agentic Large Language Model (LLM) systems decompose complex tasks into workflow Directed Acyclic Graphs (DAGs) whose primitives must be scheduled on heterogeneous clusters. Existing deep reinforcement learning (DRL) schedulers are tied to a fixed cluster size and require retraining whenever the number of servers changes. We propose SCALE (Scalable Cross-Attention Learning with Extrapolation), a DRL scheduler that generalizes to unseen cluster scales without fine-tuning. SCALE employs a cross-attention pointer network where task features query against server features, so the architecture accepts any number of servers by construction. We observe, however, that permutation-invariant architecture alone does not guarantee good performance at new scales - the attention feature undergoes distribution shift as the server count grows. To counter this, we introduce Structured Representation Regularization (SRR): a decorrelation loss combined with a KL penalty toward the standard normal, which keeps feature statistics stable regardless of input size. Trained on 16 nodes and tested directly on 32 and 48 nodes, SCALE reduces average response time by 8.9% at N=48 relative to the same architecture without SRR, confirming that explicit regularization is necessary to close the scale-generalization gap.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06786unread
Federated Foundation Models over Vehicular Networks
Kasra Borazjani, Fardis Nadimi, Payam Abdisarabshali, Owen Palinski, Allan Salihovic, Dinh Nguyen, Minghui Liwang, Seyyedali Hosseinalipour · 2026-06-08
arXiv:2606. 06786v1 Announce Type: new Abstract: This paper presents a forward-looking vision for integrating the emerging multi-modal multi-task federated foundation models (M3T FedFMs) into vehicular networks, with the goal of unifying the expressive power of multi-modal multi-task foundation models (M3T FMs) with the privacy-preserving and distributed learning capabilities of federated learning (FL).
Read next because Federated Foundation Models over Vehicular Networks 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, implement, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06786v1 Announce Type: new Abstract: This paper presents a forward-looking vision for integrating the emerging multi-modal multi-task federated foundation models (M3T FedFMs) into vehicular networks, with the goal of unifying the expressive power of multi-modal multi-task foundation models (M3T FMs) with the privacy-preserving and distributed learning capabilities of federated learning (FL). Given the largely underexplored nature of this research direction, we first introduce the fundamental training/fine-tuning principles of M3T FedFMs. We then discuss a range of their representative use cases in vehicular networks, illustrating the significant potential of M3T FedFMs to enable next-generation vehicular intelligence. Afterwards, we identify key constraints inherent to vehicular environments that challenge the practical deployment of M3T FedFMs, and articulate a set of forward-looking research directions to address these challenges. Furthermore, through a case study conducted on a real-world vehicular dataset (i.e., Waymo Open Dataset), we demonstrate the promise of M3T FedFMs for vehicular networks and release our implementation to facilitate reproducibility and stimulate research in this emerging area (repository: https://github.com/KasraBorazjani/vehicular-fedfm)
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06724unread
Synthics: Synthetic Physics-like Datasets for Machine Learning
Jari Veps\"al\"ainen · 2026-06-08
arXiv:2606. 06724v1 Announce Type: new Abstract: Representative data is fundamental in machine learning, as limited data hinders generalisation.
Read next because Synthics: Synthetic Physics-like Datasets for Machine Learning overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, compare, full, test. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06724v1 Announce Type: new Abstract: Representative data is fundamental in machine learning, as limited data hinders generalisation. Collecting sufficient real-world samples is often infeasible. Synthetic data generation offers a practical solution, but only if the generated data faithfully reflects the structure of real observations. In this paper, a method for generating synthetic regression datasets that structurally resemble physics equations from a given equation corpus is presented. The approach uses a Bayesian Probabilistic Context-Free Grammar to capture the underlying algebraic structure of the corpus, from which novel equations are sampled. To ensure the generated inputs lie within a physically meaningful domain, the applicability domain is characterised for each equation through non-intrusive probing, also recovering inter-variable constraints. Input sampling further mimics realistic experimental conditions by drawing from random sub-ranges of the valid domain with mixed uniform and truncated normal distributions. The generated data is statistically validated against the Feynman equation corpus using Kolmogorov-Smirnov tests. The generated equations match the corpus on all of the eight studied structural features, compared to only two for an unsmoothed purely probabilistic grammar, demonstrating that the Bayesian prior is essential for structural fidelity given the size of the corpus. In a downstream hyperparameter-tuning task, a gradient-boosted regressor tuned on the synthetic data picks, on average, the 6th-best configuration out of 20 on real data, matching the result of tuning on real data itself and substantially outperforming random expression trees (10th) and noise (19th).
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06718unread
MSAIC-Net: A Multi-Scale Attention and Imbalance-Aware Contrastive Network for ECG-Based Myocardial Substrate Abnormality Detection
Canyu Lei, Fenglin Zhang, Derek Bivona, Cristiane Singulane, Jonathan Pan, Kenneth Bilchick, Amit R. Patel, Jianxin Xie · 2026-06-08
arXiv:2606. 06718v1 Announce Type: new Abstract: Myocardial substrate abnormalities, such as myocardial scar and myocardial infarction (MI), are associated with adverse cardiovascular outcomes.
Read next because MSAIC-Net: A Multi-Scale Attention and Imbalance-Aware Contrastive Network for ECG-Based Myocardial Substrate Abnormality Detection 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, line, rate, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06718v1 Announce Type: new Abstract: Myocardial substrate abnormalities, such as myocardial scar and myocardial infarction (MI), are associated with adverse cardiovascular outcomes. Electrocardiography (ECG) provides a low-cost and widely available tool for detecting these abnormalities, but ECG-based detection remains challenging due to heterogeneous lead-dependent manifestations, high-dimensional multi-lead signals, class imbalance, and the limited interpretability of deep learning models. We propose a multi-scale attention-enhanced convolutional network (MSAIC-Net) for ECG-based myocardial substrate abnormality detection. MSAIC-Net employs parallel atrous convolutional branches to extract ECG features across multiple temporal receptive fields. %, enabling the model to capture both local and longer-range temporal patterns. Channel attention is then used to adaptively reweight informative lead-wise and feature-channel representations. To address class imbalance and improve feature separability, we introduce a novel imbalance-aware supervised contrastive learning strategy that encourages samples from the same class to form compact representations while increasing separation between abnormal and normal samples. Lead-wise permutation importance is further incorporated to quantify the contribution of each ECG lead and improve model interpretability. The proposed method was evaluated on two complementary datasets: a low-data institutional cohort from the University of Virginia (UVA) Health System for myocardial scar classification and the large-scale public PTB-XL dataset from PhysioNet for MI identification. Experimental results show that MSAIC-Net outperforms baseline models, with particularly pronounced improvements in the low-data UVA cohort. Overall, the proposed framework provides an effective and interpretable approach for ECG-based detection of myocardial substrate abnormalities.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06663unread
Explainable Runtime Dependency Tracking for AI-RAN Conflict Monitoring
Christie Djidjev, Nicholas Kaminski · 2026-06-08
arXiv:2606. 06663v1 Announce Type: new Abstract: Future AI-integrated Radio Access Networks (AI-RAN) will combine open programmability with learning-enabled xApps, rApps, and control functions that act on shared parameters and key performance indicators (KPIs).
Read next because Explainable Runtime Dependency Tracking for AI-RAN Conflict Monitoring 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 "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone". Matching terms: under, rate, control, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06663v1 Announce Type: new Abstract: Future AI-integrated Radio Access Networks (AI-RAN) will combine open programmability with learning-enabled xApps, rApps, and control functions that act on shared parameters and key performance indicators (KPIs). For conflict monitoring, it is not enough to know which applications are deployed; the system must also know whether the parameter--KPI dependencies assumed by runtime diagnosis remain valid under the current operating regime. This paper studies a lightweight monitoring primitive for that purpose: tracking an interpretable dependency representation from streaming telemetry events. We represent active dependencies by a Boolean matrix and use Boolean matrix multiplication to check whether recent parameter-activity and KPI-response events are consistent with the current estimate. We propose a sliding-window inference procedure that reuses the estimate when it remains consistent and recomputes it when recent observations indicate structural change. The tracker is intended as an explainable signal for conflict diagnosis and slow-loop model refresh, not as an autonomous mitigation mechanism. Experiments on controlled Boolean event streams show efficient and accurate tracking under dependency changes and Boolean observation noise.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06658unread
Capturing non-Markovian dynamics in non-equilibrium stochastic systems using flow matching
Bhargav Sriram Siddani, John B. Bell, Alejandro L. Garcia, Ishan Srivastava · 2026-06-08
arXiv:2606. 06658v1 Announce Type: new Abstract: Hydrodynamic models of stochastic particle systems represented by coarse-grained stochastic partial differential equations (SPDE), such as the regularized Dean-Kawasaki (DK) equation, do not accurately capture the short-time system dynamics that is dominated by non-Markovian effects, and low particle density regimes where the distributions are highly non-Gaussian.
Read next because Capturing non-Markovian dynamics in non-equilibrium stochastic systems using flow matching 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, line, rate, compare, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06658v1 Announce Type: new Abstract: Hydrodynamic models of stochastic particle systems represented by coarse-grained stochastic partial differential equations (SPDE), such as the regularized Dean-Kawasaki (DK) equation, do not accurately capture the short-time system dynamics that is dominated by non-Markovian effects, and low particle density regimes where the distributions are highly non-Gaussian. We develop a generative flow matching method that directly models the probability distribution of fluxes from particle simulations that explicitly incorporates non-Markovian and non-Gaussian effects. As a demonstration, we use this method to simulate the Kramers first passage time problem for a system of non-interacting Brownian particles. We show the model accurately captures the short-time behavior and provides better predictions of the statistical moments of the number density when compared against the solution of the Markovian baseline, regularized DK equation.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06647unread
The Identity Trap in EEG Foundation Models: A Diagnostic Audit
Jun-You Lin, Ying Choon Wu, Tzyy-Ping Jung · 2026-06-08
arXiv:2606. 06647v1 Announce Type: new Abstract: Objective.
Read next because The Identity Trap in EEG Foundation Models: A Diagnostic Audit overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, strong, rect, under, line, rate, without, alone. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06647v1 Announce Type: new Abstract: Objective. EEG foundation models (FMs) report strong accuracy on clinical resting-state EEG. However, high accuracy under subject-disjoint cross-validation remains ambiguous: it can reflect a genuine clinical biomarker, or subject-identity features that correlate with the label. We name this the Identity Trap and ask whether it can be diagnosed at the representation level before fine-tuning. Approach. We propose FMScope, a frozen-representation protocol packaging five diagnostics: variance decomposition, subject-axis erasure, aperiodic 1/f ablation, layer-wise label probing, and within-subject direction consistency. We apply it to three pretrained FMs (LaBraM, CBraMod, REVE) across four datasets in a 2x2 layout: subject relation of label x presence of a consensus cross-subject EEG marker. Main results. (i) The Identity Trap is universal: frozen subject-variance is 13-89x a random null in 12/12 pairs, rising in all 12 under fine-tuning (+10 to +63 pp). This dominance is a removable linear axis: erasing it improves label decoding where the label varies within subject (+6 to +12 pp in primary cells; +4 to +27 pp across external cohorts). (ii) Aperiodic 1/f is one subject carrier: removing it drops the subject probe by 9-19 pp on LaBraM and CBraMod. REVE saturates subject identity without measurable aperiodic dependence. (iii) Fine-tuning amplifies label-variance only in cells with a literature-established cross-subject marker. Significance. The Identity Trap is a physically-grounded instance of shortcut learning: the preferred cue has a measurable physiological component, and subject-disjoint splitting alone cannot rule it out. FMScope separates gains reflecting a biological marker from those reflecting subject identity.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06624unread
Principles and Practice of Deep Representation Learning: or a Mathematical Theory of Memory
San Buchanan, Druv Pai, Peng Wang, Yi Ma · 2026-06-08
arXiv:2606. 06624v1 Announce Type: new Abstract: In the current era of deep learning and especially generative models, there is significant investment in training very large generative models.
Read next because Principles and Practice of Deep Representation Learning: or a Mathematical Theory of Memory overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, control, factor, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06624v1 Announce Type: new Abstract: In the current era of deep learning and especially generative models, there is significant investment in training very large generative models. Thus far, such models have been "black boxes" that are difficult to understand in the sense that they have opaque internal mechanisms, leading to difficulties in interpretability, reliability, and control. Naturally, this lack of understanding has led to both hype and fear. This book is an attempt to "open the black box" and understand the mechanisms of large deep networks, through the perspective of representation learning, which is a major factor - arguably the single most important one - in the empirical power of deep learning models. A brief outline of this book is as follows. Chapter 1 will summarize the threads that underlie the whole text. Chapters 2, 3, 4, 5, and 6 will explain the design principles of modern neural network architectures through optimization and information theory, reducing the process of architecture development (long having been described as a sort of "alchemy") to undergraduate-level linear algebra and calculus exercises once the underlying principles are introduced. Chapters 7 and 8 will discuss applications of these principles to solve problems in more paradigmatic ways, obtaining new methods and models which are efficient, interpretable, and controllable by design, and yet no less - sometimes even more - powerful than the black-box models they resemble. Chapter 9 will discuss potential future directions for deep learning, the role of representation learning, as well as some open problems.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06572unread
Generative Models Erode Human Temporal Learning Through Market Selection
Wenjun Cao · 2026-06-08
arXiv:2606. 06572v1 Announce Type: new Abstract: We argue that modern generative models create structural risks for knowledge and cultural production at current, sub-AGI capability levels.
Read next because Generative Models Erode Human Temporal Learning Through Market Selection 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: alignment, soft, eval, source, rate, stage, capability, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06572v1 Announce Type: new Abstract: We argue that modern generative models create structural risks for knowledge and cultural production at current, sub-AGI capability levels. We define Human Temporal Learning (HTL) as path-dependent knowledge accumulation through sustained engagement with problems over time. Generative outputs increasingly resemble HTL-intensive work in surface features, so verifying whether a given output reflects genuine human learning grows costly relative to its expected benefit. Once verification loses economic justification, evaluators reward outputs regardless of production mode, and producers who invested years of learning compete on price against outputs that cost almost nothing to generate. We call this pathway value collapse and formalize it through a costly-inspection framework. Cross-domain evidence from academic publishing, legal practice, content platforms, and software security maps onto four stages of verification erosion. Alignment success is orthogonal. Better-aligned models narrow observable gaps between human and AI outputs, making source verification harder and intensifying competitive pressure against HTL-intensive work even when individual AI outputs improve.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06567unread
Are you sure? A Comprehensive and Comprehensible Survey of Uncertainty Quantification in Symbolic Regression
Julia Reuter, Fabricio Olivetti de Franca · 2026-06-08
arXiv:2606. 06567v1 Announce Type: new Abstract: Symbolic regression (SR) is a class of methods that systematically explore the space of mathematical functions to discover models that accurately capture the underlying relationships in a dataset.
Read next because Are you sure? A Comprehensive and Comprehensible Survey of Uncertainty Quantification in Symbolic Regression 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, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06567v1 Announce Type: new Abstract: Symbolic regression (SR) is a class of methods that systematically explore the space of mathematical functions to discover models that accurately capture the underlying relationships in a dataset. Despite recent advances in the field, a lack of support for uncertainty quantification (UQ) limits its adoption in real-world decision processes. In regression analysis, UQ provides important information about the model reliability, which can both help to avoid overfitting by accounting for uncertainty in the data, and provide insights for decision-making. This survey is the first to clearly address this issue, with the objective of introducing essential UQ concepts and reviewing the current literature on UQ in SR, which can be broadly organized into three research directions: frequentist, Bayesian, and model selection. Despite its importance, UQ in SR is still underexplored, which motivates further research into reliable UQ methods for SR.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06554unread
Multi-Scale Feature Attention Network for Polymer Classification using THz Dual-Comb Spectroscopy
Roshni Mahtani, Il\'an Carretero, Laura Monroy, Aldo Moreno-Oyervides, Oscar El\'ias Bonilla-Manrique, Roc\'io del Amor · 2026-06-08
arXiv:2606. 06554v1 Announce Type: new Abstract: Reliable polymer identification is essential for ensuring the quality and safety of recycled plastics, yet conventional sorting and spectroscopic techniques often struggle to deliver robust discrimination.
Read next because Multi-Scale Feature Attention Network for Polymer Classification using THz Dual-Comb Spectroscopy 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, rate, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06554v1 Announce Type: new Abstract: Reliable polymer identification is essential for ensuring the quality and safety of recycled plastics, yet conventional sorting and spectroscopic techniques often struggle to deliver robust discrimination. Terahertz Dual-Comb Spectroscopy (THz-DCS) offers a promising alternative, providing rapid, high-resolution, and non-destructive measurements. In this work, we leverage THz-DCS to classify 12 types of polymers, including pure polymers, multilayer films, commercial blends, and biopolymers. To handle the complexity of these spectral signals, we propose the Multi-Scale Feature Attention Network (MSFAN), a novel deep learning architecture tailored for THz-DCS data. The framework integrates feature gating for signal recalibration and multi-scale parallel convolutions to capture diverse frequency patterns. These features are further refined through cross-feature attention and attention pooling, enabling the model to intrinsically highlight the most informative THz regions. MSFAN consistently outperforms state-of-the-art models, reaching a classification accuracy of 85.2%. This study demonstrates the potential of combining THz-DCS with deep learning techniques for effective, scalable, and interpretable polymer classification.
- score 100arxiv stat.ML (Machine Learning)arxiv:2404.02141unread
Robustly estimating heterogeneity in factorial data using Rashomon Partitions
Aparajithan Venkateswaran, Anirudh Sankar, Arun G. Chandrasekhar, Tyler H. McCormick · 2026-06-08
arXiv:2404. 02141v5 Announce Type: replace-cross Abstract: In both observational data and randomized control trials, researchers select statistical models to articulate how the outcome of interest varies with combinations of observable covariates.
Read next because Robustly estimating heterogeneity in factorial data using Rashomon Partitions overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, rate, control, without, factor, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2404.02141v5 Announce Type: replace-cross Abstract: In both observational data and randomized control trials, researchers select statistical models to articulate how the outcome of interest varies with combinations of observable covariates. Choosing a model that is too simple can obfuscate important heterogeneity in outcomes between covariate groups, while too much complexity risks identifying spurious patterns. In this paper, we propose a novel Bayesian framework for model uncertainty called Rashomon Partition Sets (RPSs). The RPS consists of all models that have posterior density close to the maximum a posteriori (MAP) model. We construct the RPS by enumeration, rather than sampling, which ensures that we explore all models with high evidence in the data, even if they offer dramatically different substantive explanations. We use a l0 prior, which allows the allows us to capture complex heterogeneity without imposing strong assumptions about the associations between effects, showing this prior is minimax optimal from an information-theoretic perspective. We characterize the approximation error of (functions of) parameters computed conditional on being in the RPS relative to the entire posterior. We propose an algorithm to enumerate the RPS from the class of models that are interpretable and unique, then provide bounds on the size of the RPS. We give simulation evidence along with three empirical examples: price effects on charitable giving, heterogeneity in chromosomal structure, and the introduction of microfinance.
- score 100arxiv stat.ML (Machine Learning)arxiv:2109.02644unread
Resolvent convergence for sample covariance matrices with general covariance profiles and quadratic-form control
Cosme Louart · 2026-06-08
arXiv:2109. 02644v5 Announce Type: replace-cross Abstract: We study the resolvent \[ G^z = \left(\frac{1}{n}XX^T - zI_p\right)^{-1}, \qquad z\in\mathbb C,\ \Im(z)>0, \] where $X=(x_1,\ldots,x_n)\in\mathcal M_{p,n}$ is a random matrix with independent, but not necessarily identically distributed, columns.
Read next because Resolvent convergence for sample covariance matrices with general covariance profiles and quadratic-form control overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, control. Source: arxiv stat.ML (Machine Learning).
arXiv:2109.02644v5 Announce Type: replace-cross Abstract: We study the resolvent \[ G^z = \left(\frac{1}{n}XX^T - zI_p\right)^{-1}, \qquad z\in\mathbb C,\ \Im(z)>0, \] where $X=(x_1,\ldots,x_n)\in\mathcal M_{p,n}$ is a random matrix with independent, but not necessarily identically distributed, columns. Our bounds are expressed in terms of moments of the centered quadratic forms \[ q_i(A):=x_i^TAx_i-\mathbb E[x_i^TAx_i], \] for deterministic matrices $A$ with unit Hilbert--Schmidt norm. In particular, we do not assume independence between the entries of a given column $x_i$. In the quasi-asymptotic regime $p\le O(n)$, the matrix $G^z$ admits a natural deterministic equivalent $\tilde G^z$, depending only on the second moments of the column vectors $x_1,\ldots,x_n$. We show that, for any deterministic matrix $B\in\mathcal M_p$, the trace $\text{Tr}(BG^z)$ is close to $\text{Tr}(B\tilde G^z)$, with error controlled by $\|B\|_{\text{HS}}$ under first-moment bounds on the quadratic forms, and by $\|B\|_{\text{HS}}/\sqrt n$ under suitable second-moment bounds.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.05967unread
Fast and Robust Convergence Rate for TD(0) with Linear Function Approximation, Universal Learning Steps and I.I.D. Samples
Ziad Kobeissi (L2S), \'Elo\"ise Berthier (U2IS) · 2026-06-08
arXiv:2606. 05967v2 Announce Type: replace Abstract: In this paper, we study the finite-time behavior of the TD(0) temporal-difference method with linear function approximation (LFA).
Read next because Fast and Robust Convergence Rate for TD(0) with Linear Function Approximation, Universal Learning Steps and I.I.D. Samples overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, rate, does, chain, on-policy, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.05967v2 Announce Type: replace Abstract: In this paper, we study the finite-time behavior of the TD(0) temporal-difference method with linear function approximation (LFA). We consider on-policy independent and identically distributed (i.i.d.) samples, a constant learning step, and the Polyak-Juditsky averaging method. We establish a new convergence rate, for the Mean-Square Error (MSE) on the approximated function, that is (i) fast in the sense that it admits an optimal dependency in the number of iterations k (i.e., of order 1/k), (ii) robust to ill-conditioning: it only depends on an initial error and modelindependent constants and (iii) sharp up to a multiplicative constant lower than 11. In particular, it does not depend on the smallest eigenvalue of the uncentered covariance matrix of the linear parametrization, unlike all pre-existing O(1/k) rates in the TD(0) literature. We also introduce PCTD(0), a variant of TD(0), which benefits from better convergence properties under an additional assumption of strong mixing on the Markov Chain.
- score 100arxiv stat.ML (Machine Learning)arxiv:2603.20967unread
Hard labels sampled from sparse targets mislead rotation invariant algorithms
Avrajit Ghosh, Bin Yu, Manfred Warmuth, Peter Bartlett · 2026-06-08
arXiv:2603. 20967v2 Announce Type: replace Abstract: One of the most common machine learning setups is logistic regression.
Read next because Hard labels sampled from sparse targets mislead rotation invariant algorithms 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, soft, line, trained, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2603.20967v2 Announce Type: replace Abstract: One of the most common machine learning setups is logistic regression. In many classification models, including neural networks, the final prediction is obtained by applying a logistic link function to a linear score. In binary logistic regression, the feedback can be either soft labels, corresponding to the true conditional probability of the data (as in distillation), or sampled hard labels (taking values $\pm 1$). We point out a fundamental problem that arises even in a particularly favorable setting, where the goal is to learn a noise-free soft target of the form $\sigma(\mathbf{x}^{\top}\mathbf{w}^{\star})$. In the over-constrained case (i.e. the number of samples $n$ exceeds the input dimension $d$) with examples $(\mathbf{x}_i,\sigma(\mathbf{x}_i^{\top}\mathbf{w}^{\star}))$, it is sufficient to recover $\mathbf{w}^{\star}$ and hence achieve the Bayes risk. However, we prove that when the examples are labeled by hard labels $y_i$ sampled from the same conditional distribution $\sigma(\mathbf{x}_i^{\top}\mathbf{w}^{\star})$ and $\mathbf{w}^{\star}$ is $s$-sparse, then rotation-invariant algorithms are provably suboptimal: they incur an excess risk $\Omega\!\left(\frac{d-1}{n}\right)$, while there are simple non-rotation invariant algorithms with excess risk $O(\frac{s\log d}{n})$. The simplest rotation invariant algorithm is gradient descent on the logistic loss (with early stopping). A simple non-rotation-invariant algorithm for sparse targets that achieves the above upper bounds uses gradient descent on the weights $u_i,v_i$, where now the linear weight $w_i$ is reparameterized as $u_iv_i$.
- score 100arxiv stat.ML (Machine Learning)arxiv:2602.06245unread
Inheritance Between Feedforward and Convolutional Networks via Model Projection
Nicolas Ewen, Jairo Diaz-Rodriguez, Kelly Ramsay · 2026-06-08
arXiv:2602. 06245v2 Announce Type: replace Abstract: Neural-network techniques are often transferred across architecture families by analogy, but such transfer is valid only when the assumptions required by a technique are preserved.
Read next because Inheritance Between Feedforward and Convolutional Networks via Model Projection 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, line, project, full, trained, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2602.06245v2 Announce Type: replace Abstract: Neural-network techniques are often transferred across architecture families by analogy, but such transfer is valid only when the assumptions required by a technique are preserved. We introduce this idea as inheritance between model classes. Using a unified node-level framework with tensor-valued activations, we prove that generalized feedforward networks (GFFNs) form a strict subset of generalized convolutional networks (GCNNs), so GCNN properties transfer directly to GFFNs. The reverse direction is not automatic: standard CNN nodes use spatial kernels, while FFN nodes use one scalar weight per input contribution. We introduce model projection to recover a restricted reverse inheritance path. Projection freezes each convolutional input-channel sub-function and learns one scalar coefficient for each input-output channel contribution, giving projected CNN nodes the GFFN-style trainable structure of scalar-weighted input recombination. This inherited structure leads naturally to parameter-efficient transfer learning. Across multiple ImageNet-pretrained CNN backbones and downstream image-classification datasets, model projection is competitive with standard and PEFT baselines and provides an effective initialization for subsequent full fine-tuning.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.07392unread
Online Pandora's Box for Contextual LLM Cascading
Alexandre Belloni, Yan Chen, Yehua Wei · 2026-06-08
arXiv:2606. 07392v1 Announce Type: cross Abstract: Motivated by Large Language Model (LLM) cascading, we propose an online contextual Pandora's Box model for adaptively querying and selecting LLM APIs.
Read next because Online Pandora's Box for Contextual LLM Cascading overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, under, eval, line, rate, cascading. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.07392v1 Announce Type: cross Abstract: Motivated by Large Language Model (LLM) cascading, we propose an online contextual Pandora's Box model for adaptively querying and selecting LLM APIs. In each period, a decision-maker observes a request context and faces a two-phase decision problem. In the query phase, the decision-maker sequentially queries APIs, where each query reveals a generated output and the decision-maker incurs an (output-dependent) cost. In the selection phase, the decision-maker selects one of the generated outputs to deploy and observes only the downstream reward of the deployed output. This output-mediated feedback structure differs from classical online contextual Pandora's Box models, in which opening a box directly reveals its reward. Rather than estimating the full conditional output and cost distributions of each API, we directly model the reservation index and develop a learning approach for the query phase. Specifically, we impose a parametric structure on the contextual reservation index functions induced by the classical Weitzman's policy. Our policy combines generalized method of moments (GMM) type estimation of these reservation indices with UCB-style confidence bounds for both these indices and the shared output-level reward evaluator. Under regularity conditions, we prove that the resulting policy achieves dimension-dependent $\widetilde O(\sqrt T)$ cumulative regret over a horizon of $T$ periods.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.07247unread
Theory of learning of high-dimensional controlled non-linear dynamical systems (I): models and methods
Pierfrancesco Urbani · 2026-06-08
arXiv:2606. 07247v1 Announce Type: cross Abstract: Neural ordinary differential equations (neural ODEs) have rapidly gained prominence as a powerful and unifying framework for conceptualizing artificial neural networks, elegantly connecting the continuous-time modeling of dynamical systems with the discrete, data-driven paradigm of modern deep learning.
Read next because Theory of learning of high-dimensional controlled non-linear dynamical systems (I): models and methods 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, token, line, control, trained, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.07247v1 Announce Type: cross Abstract: Neural ordinary differential equations (neural ODEs) have rapidly gained prominence as a powerful and unifying framework for conceptualizing artificial neural networks, elegantly connecting the continuous-time modeling of dynamical systems with the discrete, data-driven paradigm of modern deep learning. Beyond their practical advantages they offer fresh theoretical insights into the training and generalization properties of neural networks. The distinctive feature of this framework is its dual dynamical nature: inference dynamics, which govern the ODE evolution during forward computation, and training dynamics, which control the optimization of model parameters. This makes neural ODEs a particularly well-suited theoretical framework for studying a large variety of settings such as multi-layer neural networks (ResNets for example), autoregressive models (with next-token generation dynamics), generative models, and recurrent neural networks in theoretical neuroscience. In this work, we introduce a theoretically grounded class of models for studying neural ODEs trained via online stochastic gradient descent. We solve the training dynamics of these models via dynamical mean field theory and derive learning curves in the high-dimensional limit.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.07124unread
Information-Theoretic Bounds for Sparse Covariance Estimation in the Vertical-Split Distributed Model
Jing Yee Tan, Guangyue Han · 2026-06-08
arXiv:2606. 07124v1 Announce Type: cross Abstract: We study the minimax estimation error for distributed covariance matrix estimation in the vertical-split (feature-split) setting, where two agents each observe different coordinates of $m$ i.
Read next because Information-Theoretic Bounds for Sparse Covariance Estimation in the Vertical-Split Distributed Model overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, compare, does, factor, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.07124v1 Announce Type: cross Abstract: We study the minimax estimation error for distributed covariance matrix estimation in the vertical-split (feature-split) setting, where two agents each observe different coordinates of $m$ i.i.d. sub-Gaussian samples and communicate a limited number of bits to a central server. While Rahmani et al. [2025] established nearly tight bounds for dense (unstructured) cross-covariance matrices, we investigate whether imposing elementwise $s$-sparsity on the cross-covariance $C_{21}$ can reduce the required communication and sample complexity. In contrast to the horizontal-split setting, where Braverman et al. [2016] showed that sparsity does not reduce communication cost for mean estimation, we prove that sparsity does help for cross-covariance estimation in the vertical split. Specifically, we establish minimax lower bounds showing that the communication budget per agent scales as $B_k = \Omega(\sigma^4 d_k\, s' \log(d_1 d_2/s')/\varepsilon^2)$ and the sample complexity for cross-covariance estimation as $m = \Omega(\sigma^4\, s' \log(d_1 d_2/s')/\varepsilon^2)$, where $s' = s \wedge d_{\min}$. For the $1$-sparse case, this yields an exponential improvement from $d_1 d_2$ to $\log(d_1 d_2)$ compared to the dense rate. Our lower bounds are established via Fano's method with an explicit sparse packing using a Varshamov--Gilbert-type argument for signed partial permutation matrices combined with the Conditional Strong Data Processing Inequality of Rahmani et al. [2025]. We show the bounds are tight with a matching achievable scheme, based on covering-net quantization and entry-wise hard thresholding, that attains the $s$-sparse lower bound up to polylogarithmic factors.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.07058unread
Constructing VAE Latent Spaces with Prescribed Topology
Jilles S. van Hulst, Jakub M. Tomczak, W. P. M. H. Heemels, Duarte J. Antunes · 2026-06-08
arXiv:2606. 07058v1 Announce Type: cross Abstract: Variational autoencoders (VAEs) learn low-dimensional latent representations of high-dimensional data.
Read next because Constructing VAE Latent Spaces with Prescribed Topology overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, circle, soft, line, project, factor, symmetry. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.07058v1 Announce Type: cross Abstract: Variational autoencoders (VAEs) learn low-dimensional latent representations of high-dimensional data. When the data lies on a manifold with non-Euclidean topology, the standard Gaussian prior introduces a topological mismatch that degrades reconstruction quality and prevents faithful representation. We present a constructive mathematical framework that resolves this mismatch for all manifolds that admit a product covering space. These are manifolds expressible as products of elementary factors (circles, intervals, or lines) or as quotients of such products by a finite symmetry group. The class includes cylinders, tori, M\"{o}bius strips, Klein bottles, and real projective spaces. Factorized distributions over the elementary factors yield product topologies with closed-form, decoupled KL divergences, so that each latent factor can be shaped independently while keeping training tractable. We catalogue reparametrizable encoder-prior pairs for periodic, bounded, and unbounded supports, and provide coordinate transformations that allow standard neural networks to output non-Euclidean parameters with smooth gradients. For quotient manifolds, the decoder receives group-invariant features of the covering-space coordinates, so that identified points produce identical outputs. Anchor constraints fix the coordinate system relative to the data or create soft topological holes. Experiments on synthetic manifolds and real-image datasets (rotated and cyclically shifted MNIST) confirm that a topology-matched prior aligns KL regularization with the data manifold. The resulting topology-aware models outperform the Gaussian baseline at all practically relevant regularization strengths. The code is available at https://github.com/JvHulst/VAE-Topology.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.06576unread
Gaussian Process Latent Factor Regression for Low-Data, High-Dimensional Output Problems
Edward T. Stevenson, Eric T. Wolf, Mei Ting Mak, N. J. Mayne, Miles Cranmer · 2026-06-08
arXiv:2606. 06576v1 Announce Type: new Abstract: In the sciences, regression tasks often require predicting high-dimensional outputs from few training examples.
Read next because Gaussian Process Latent Factor Regression for Low-Data, High-Dimensional Output Problems overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, rate, factor, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06576v1 Announce Type: new Abstract: In the sciences, regression tasks often require predicting high-dimensional outputs from few training examples. Multi-output Gaussian processes excel in low-data regimes but typically struggle with high-dimensional outputs. Compress-then-predict pipelines such as PCA-GP (principal component analysis plus Gaussian process regression) handle high dimensionality, but rely on bases optimized for reconstruction rather than prediction. To address this gap, we propose a model that represents each output as a linear-Gaussian decoding of a low-dimensional latent state drawn from a Gaussian process prior. By analytically marginalizing the decoder weights, we couple compression and prediction in a single objective that scales to high-dimensional outputs. We refer to this model as Gaussian process latent factor regression (GPLFR). We demonstrate GPLFR by building the first spatially resolved emulator of global climate models for rocky exoplanets.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.06957unread
Deep Single-Index Fr\'echet Regression
Muqing Cui, Yidong Zhou, Su I Iao, Hans-Georg M\"uller · 2026-06-08
arXiv:2606. 06957v1 Announce Type: new Abstract: Predicting outputs that are located in non-Euclidean spaces, such as probability distributions, networks, and symmetric positive-definite matrices, is becoming increasingly important in modern data analysis, particularly when inputs are high-dimensional.
Read next because Deep Single-Index Fr\'echet Regression overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, position. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.06957v1 Announce Type: new Abstract: Predicting outputs that are located in non-Euclidean spaces, such as probability distributions, networks, and symmetric positive-definite matrices, is becoming increasingly important in modern data analysis, particularly when inputs are high-dimensional. We propose DeSI (Deep Single-Index Fr\'echet Regression), a semiparametric framework for regression with metric space-valued outputs and multivariate inputs that assumes a single-index structure for the conditional Fr\'echet mean. DeSI estimates an interpretable index direction, which quantifies the relative importance of inputs, using a deep neural network, and performs Fr\'echet regression along the resulting one-dimensional index in the target metric space. This structure mitigates the curse of dimensionality while retaining interpretability, which stands in contrast to standard deep neural networks. We establish theoretical guarantees for DeSI, including uniform approximation and convergence rates, and demonstrate its strong predictive performance through simulations on distributions, networks, and symmetric positive-definite matrices, as well as an application to compositional mood data from New Jersey.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.06814unread
The Effect of Training Task Diversity on In-Context Learning through the Lens of Low-Dimensional Subspaces
Soo Min Kwon, Alec S. Xu, Can Yaras, Dogyoon Song, Laura Balzano, Qing Qu · 2026-06-08
arXiv:2606. 06814v1 Announce Type: new Abstract: The transformer's emergent ability to perform in-context learning (ICL) has sparked a wide range of studies designed to understand its underlying mechanisms.
Read next because The Effect of Training Task Diversity on In-Context Learning through the Lens of Low-Dimensional Subspaces overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.06814v1 Announce Type: new Abstract: The transformer's emergent ability to perform in-context learning (ICL) has sparked a wide range of studies designed to understand its underlying mechanisms. Existing works often study how training task diversity, defined either as the number of ICL training task vectors or as the number of function classes from which the task vectors are drawn, shapes both the learning dynamics and generalization capabilities of ICL. While both definitions have uncovered many interesting phenomena, many observations under the latter definition remain theoretically unexplained. This paper presents a minimal analytical model under which these phenomena provably emerge from the properties of the training data. By modeling the training task vectors as a mixture of low-rank Gaussians, we show how training task diversity, defined by the number of non-overlapping columns between subspaces that parameterize the covariance matrices, improves both the generalization and optimization trajectory of ICL with linear attention. In particular, we show that our model can explain (i) why training with task diversity shortens the ICL plateau and (ii) why ICL appears to achieve out-of-distribution generalization. We conclude by empirically demonstrating how our results extend to nonlinear transformers and nonlinear function classes. Overall, our work presents a tractable framework to unify existing observations.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.06772unread
Generalization in Deep Neural Networks: Minimax Rates for Gradient Methods
Junyu Zhou, Puyu Wang, Yunwen Lei, Marius Kloft, Yiming Ying · 2026-06-08
arXiv:2606. 06772v1 Announce Type: new Abstract: Understanding the generalization performance of over-parameterized neural networks has become a central topic in deep learning theory.
Read next because Generalization in Deep Neural Networks: Minimax Rates for Gradient Methods 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, width, rate, full, trained. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.06772v1 Announce Type: new Abstract: Understanding the generalization performance of over-parameterized neural networks has become a central topic in deep learning theory. While recent advances, particularly works under the Neural Tangent Kernel (NTK) regime, have shed light on the behavior of shallow architectures, the statistical generalization properties of deep neural networks (DNNs), especially in regression tasks, remain far less understood. In this paper, we make significant progress toward closing this gap by providing a comprehensive generalization analysis of DNNs trained using gradient-based methods. First, we establish, for the first time, a crucial connection between the learning dynamics of a DNN with smooth activation functions trained via gradient-based methods and those of kernel methods, showing that gradient-based methods on over-parameterized DNNs can fully inherit the favorable learning dynamics of their kernel counterparts. Building on this connection and the well-established optimality of kernel methods, we derive the first known minimax-optimal rates for the excess population risk of both gradient descent (GD) and stochastic gradient descent (SGD), under the assumption that network width scales polynomially with the sample size. Our results demonstrate that, with sufficient width, DNNs trained by GD or SGD can achieve generalization performance comparable to kernel-based methods.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.06764unread
Optimal Rates for Generalization of Gradient Descent Methods with Deep Neural Networks
Junyu Zhou, Puyu Wang, Yunwen Lei, Yiming Ying, Ding-Xuan Zhou · 2026-06-08
arXiv:2606. 06764v1 Announce Type: new Abstract: Recent progress has been made in understanding the statistical generalization performance of gradient descent methods for overparameterized neural networks within the neural tangent kernel (NTK) regime.
Read next because Optimal Rates for Generalization of Gradient Descent Methods with Deep Neural Networks 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, width, rate, trained. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.06764v1 Announce Type: new Abstract: Recent progress has been made in understanding the statistical generalization performance of gradient descent methods for overparameterized neural networks within the neural tangent kernel (NTK) regime. However, most of the existing work on regression problems is limited to shallow network architectures, leaving a notable gap in the theory of deep neural networks. This paper addresses this gap by presenting a comprehensive generalization analysis for deep ReLU networks trained using gradient descent (GD) and stochastic gradient descent (SGD). Specifically, we establish the first known minimax-optimal rates of excess population risk for both GD and SGD with deep ReLU networks, under the assumption that the network width scales polynomially with respect to the network depth and training sample size. Our results demonstrate that with sufficient width, gradient descent methods for deep ReLU networks can achieve optimal generalization rates on par with kernel methods.
- score 94arxiv cs.CR (Cryptography and Security)arxiv:2606.07282unread
Rethinking IoT Intrusion Detection: Augmenting Routing Metrics with Radio Features
Yichang Sun, Andreas Johnsson, Sourasekhar Banerjee · 2026-06-08
arXiv:2606. 07282v1 Announce Type: new Abstract: Machine learning-based intrusion detection systems (IDS) for RPL-based IoT networks often rely solely on routing layer features, which provide only a partial view of network behaviour.
Read next because Rethinking IoT Intrusion Detection: Augmenting Routing Metrics with Radio Features 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 "Implement Chen et al. persona-vector extraction recipe and compare to project's centroid-difference recipe". Matching terms: under, eval, compare, alone. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07282v1 Announce Type: new Abstract: Machine learning-based intrusion detection systems (IDS) for RPL-based IoT networks often rely solely on routing layer features, which provide only a partial view of network behaviour. In this work, we investigate whether incorporating Transmit (TX) and Receive (RX) radio features alongside the standard RPL feature set can improve detection performance in an LSTM-based IDS. We evaluate the proposed approach across three different attack types, namely DIS-Flooding, Local Repair, and Worst Parent under varying network sizes. The results show that incorporating TX and RX improves the IDS's overall detection performance by up to ~4% in F1-score compared with using routing-layer features alone, with the most notable gain observed for the Worst Parent attack.
- score 94arxiv stat.ML (Machine Learning)arxiv:2606.05919unread
Finding Most Influential Sets
Lucas D. Konrad, Nikolas Kuschnig · 2026-06-08
arXiv:2606. 05919v2 Announce Type: replace Abstract: Identifying most influential sets (MIS) - size-$k$ subsets whose removal maximally changes a target estimand - is typically infeasible because it requires searching over $\binom{n}{k}$ subsets.
Read next because Finding Most Influential Sets 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 "#351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation". Matching terms: under, line, rate, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.05919v2 Announce Type: replace Abstract: Identifying most influential sets (MIS) - size-$k$ subsets whose removal maximally changes a target estimand - is typically infeasible because it requires searching over $\binom{n}{k}$ subsets. For estimands with linear-fractional leave-set-out effects, we show that MIS selection reduces to a one-parameter sequence of top-$k$ problems. Dinkelbach's method yields an algorithm with $\mathcal{O}(n)$ cost per iteration and finite termination. For fixed residualized inputs, the algorithm returns a globally optimal set for the univariate ratio objective, including the oracle-residualized partial linear model. With estimated nuisance functions, uniform denominator and generated-score stability imply approximation to the first-order oracle orthogonal-score objective; exact set recovery follows under a separation condition. Simulations and applications show that the method recovers exact MIS that were previously computationally inaccessible.
- score 94arxiv stat.ML (Machine Learning)arxiv:2604.03146unread
Characterization of Gaussian Universality Breakdown in High-Dimensional Empirical Risk Minimization
Chiheb Yaakoubi, Cosme Louart, Malik Tiomoko, Zhenyu Liao · 2026-06-19
arXiv:2604. 03146v3 Announce Type: replace Abstract: We study high-dimensional convex empirical risk minimization (ERM) under general non-Gaussian data designs.
Read next because Characterization of Gaussian Universality Breakdown in High-Dimensional Empirical Risk Minimization 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 "Implement Chen et al. persona-vector extraction recipe and compare to project's centroid-difference recipe", experiment "Test FR↔IT bystander-spill symmetry at multi-seed + 5 phrasings — pooled-rate vs per-phrasing asymmetry from #239 fact-check". Matching terms: under, project, test, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2604.03146v3 Announce Type: replace Abstract: We study high-dimensional convex empirical risk minimization (ERM) under general non-Gaussian data designs. By heuristically extending the Convex Gaussian Min-Max Theorem (CGMT) to non-Gaussian settings, we derive an asymptotic min-max characterization of key statistics, enabling approximation of the mean $\mu_{\hat{\theta}}$ and covariance $C_{\hat{\theta}}$ of the ERM estimator $\hat{\theta}$. Specifically, under a concentration assumption on the data matrix and standard regularity conditions on the loss and regularizer, we show that for a test covariate $x$ independent of the training data, the projection $\hat{\theta}^\top x$ approximately follows the convolution of the generally non-Gaussian distribution of $\mu_{\hat{\theta}}^\top x$ with an independent centered Gaussian variable of variance $\mathrm{tr}(C_{\hat{\theta}} \mathbb{E}[xx^\top])$. This result clarifies the scope and limits of Gaussian universality for ERMs. Additionally, we prove that any $\mathcal{C}^2$ regularizer is asymptotically equivalent to a quadratic form determined solely by its Hessian at zero and gradient at $\mu_{\hat{\theta}}$. Numerical simulations across diverse losses and models are provided to validate our theoretical predictions and qualitative insights.
- score 90arxiv stat.ML (Machine Learning)arxiv:2206.08598unread
Characterizing Learning Dynamics under Relative Reparameterization of Singular Models
Pascal Mattia Esser, Frank Nielsen · 2026-06-08
arXiv:2206. 08598v2 Announce Type: replace-cross Abstract: A common way to analyze learning of statistical models is to consider operations in the models parameter space, however this becomes challenging when there is no one-to-one mapping between the parameter space and the underlying statistical model space.
Read next because Characterizing Learning Dynamics under Relative Reparameterization of Singular 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 "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". Matching terms: under, rate, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2206.08598v2 Announce Type: replace-cross Abstract: A common way to analyze learning of statistical models is to consider operations in the models parameter space, however this becomes challenging when there is no one-to-one mapping between the parameter space and the underlying statistical model space. Such ``singular models'' occur frequently and exhibit a characteristic decrease in convergence speed of learning trajectories due to attractor behaviors. In this work, we consider a relative reparameterization technique of the parameter space, which yields a general method for extracting regular sub-models from singular models. On the example of Gaussian Mixture Models and Neural Networks we theoretically and numerically analyze the convergence rate for Gradient Descent under both parameterizations. Analyzing second-order methods and explicit properties of the Fisher Information Matrix we distinguish between differences in convergence behavior arising from algorithmic and intrinsic information-geometric aspects.
- score 86arxiv stat.ML (Machine Learning)arxiv:2606.06855unread
Stability beyond Bounded Differences: Sharp Generalization Bounds under Finite $L_p$ Moments
Qianqian Lei, Soham Bonnerjee, Yuefeng Han, Wei Biao Wu · 2026-06-08
arXiv:2606. 06855v1 Announce Type: new Abstract: While algorithmic stability is a central tool for understanding generalization of learning algorithms, existing high-probability guarantees typically rely on uniform boundedness or sub-Gaussian/sub-Weibull tail assumptions, which can be overly restrictive for modern settings with heavy-tailed or unbounded losses.
Read next because Stability beyond Bounded Differences: Sharp Generalization Bounds under Finite $L_p$ Moments 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. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.06855v1 Announce Type: new Abstract: While algorithmic stability is a central tool for understanding generalization of learning algorithms, existing high-probability guarantees typically rely on uniform boundedness or sub-Gaussian/sub-Weibull tail assumptions, which can be overly restrictive for modern settings with heavy-tailed or unbounded losses. We develop a stability-based framework that requires only a finite $L_p$ moment condition. Our first contribution is sharp concentration inequalities for functions of independent random variables under $L_p$ constraints, extending McDiarmid's bounded-differences techniques beyond the classical regime. Leveraging these results, we derive sharp high-probability generalization bounds across a range of learning paradigms, including empirical risk minimization, transductive regression, and meta-learning. These guarantees show that $L_p$ stability suffices for robust generalization even when boundedness fails, substantially weakening the standard assumptions in the stability literature.
- score 78arxiv stat.ML (Machine Learning)arxiv:2606.06782unread
The Sharp Phase Transition of Tyler's M-Estimator for Robust Subspace Recovery
Gilad Lerman, Teng Zhang · 2026-06-08
arXiv:2606. 06782v1 Announce Type: cross Abstract: Robust Subspace Recovery (RSR) aims to identify an underlying d-dimensional subspace from a dataset heavily corrupted by outliers.
Read next because The Sharp Phase Transition of Tyler's M-Estimator for Robust Subspace Recovery 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 "#351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation". Matching terms: under, rate, position. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.06782v1 Announce Type: cross Abstract: Robust Subspace Recovery (RSR) aims to identify an underlying d-dimensional subspace from a dataset heavily corrupted by outliers. Complexity-theoretic results establish a threshold for the problem's computational hardness based on the dimension-scaled signal-to-noise ratio (DS-SNR): the problem is SSE-hard when the DS-SNR is strictly less than 1, and solvable via practical algorithms when it is greater than 1 under general position assumptions. However, the exact behavior of practical algorithms at the critical boundary DS-SNR = 1 has remained unknown. This work resolves the behavior of Tyler's M-estimator (TME) at this critical boundary, consequently establishing a sharp phase transition. Specifically, we prove that TME converges exactly to the true subspace for DS-SNR \geq 1 under a new stability condition, which is less restrictive than the general position assumptions used in prior literature. Our analysis utilizes a decomposition of the TME iterates within a majorization-minimization framework.
- score 58arxiv cs.AI (Artificial Intelligence)arxiv:2606.06656unread
A Study of Parallel Continuous Local Search
Cody J Christopher, Charles Gretton · 2026-06-08
arXiv:2606. 06656v1 Announce Type: new Abstract: We study parallel Continuous Local Search (CLS) as a solution approach for Boolean satisfiability problems with symmetric pseudo-Boolean (PB) constraints.
Read next because A Study of Parallel Continuous Local Search 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". Matching terms: rate. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06656v1 Announce Type: new Abstract: We study parallel Continuous Local Search (CLS) as a solution approach for Boolean satisfiability problems with symmetric pseudo-Boolean (PB) constraints. Here, the $n$-variable PB-satisfiability problem is relaxed to a continuous optimisation problem with a differentiable objective function on an $n$-dimensional hypercube. For satisfiable instances, the global minimisers of this optimisation problem correspond to satisfying assignments of the SAT problem at hand. We present several novel findings via empirical experiments: (i) redundant constraints can inhibit rather than accelerate convergence; (ii) CLS shows promise as a sub-solver in hybridised settings, quickly completing partial assignments; and (iii) local search rapidly converges to a stable distribution of solution quality (i.e., degree of satisfaction), due to saddle-dense objectives where additional solver steps yield diminishing returns. Our findings inform practical uses of CLS for SAT on modern accelerator hardware.
- score 58arxiv cs.LG (Machine Learning)arxiv:2606.06722unread
Flatland: The Adventures of Gradient Descent with Large Step Sizes
Leonardo Galli, Curtis Fox, Wiebke Bartolomaeus, Mark Schmidt, Holger Rauhut · 2026-06-08
arXiv:2606. 06722v1 Announce Type: new Abstract: The training of neural networks often entails objective functions that are not globally $L$-smooth.
Read next because Flatland: The Adventures of Gradient Descent with Large Step Sizes 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". Matching terms: rate. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06722v1 Announce Type: new Abstract: The training of neural networks often entails objective functions that are not globally $L$-smooth. For these functions, it is both theoretically and practically difficult to reply to the question: what is the largest possible step size that ensures the convergence of gradient descent (GD)? We address this longstanding open question in deep learning by providing a unifying definition of "large" step sizes that requires only local Lipschitz (or even H\"older) continuity of the gradient. We design first-order adaptive methods that provably yield large step sizes and show that they operate at the edge of stability (EoS) right from the start of the training. In particular, the loss decreases nonmonotonically and the product between the step size and sharpness, i.e., the largest eigenvalue of the Hessian, stays above the EoS threshold of 2 throughout training. Using our method, we are also able to minimize the sharpness all the way down to its global minimum. Contrary to expectation, we find that encountering globally-flat regions too early in the training may both slow down convergence and jeopardize the generalization ability of the network. Exploiting a self-stabilization argument, we allow GD to enter slightly sharper valleys and turn unsuccessful training runs into very successful ones.
New research
- score 30arxiv stat.ML (Machine Learning)arxiv:2606.07213unread
Principal Component Analysis for Multivariate Extremes
Dan Cooley, Anne Sabourin, Troy Wixson · 2026-06-08
arXiv:2606. 07213v1 Announce Type: cross Abstract: This chapter explores ways to reduce the dimensionality of the data while preserving key information relevant to the analysis of multivariate extreme values.
Background read from arxiv stat.ML (Machine Learning). It did not strongly match recent Sagan clean results, beliefs, or experiments, so keep it lower priority unless the title is independently relevant.
arXiv:2606.07213v1 Announce Type: cross Abstract: This chapter explores ways to reduce the dimensionality of the data while preserving key information relevant to the analysis of multivariate extreme values.
Threats and caveats
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06833unread
Hearing the Unspoken: Language Model Priors for Acoustic Adversarial Attacks
Jiani Xie, Andrew C. Cullen, Paul Montague, Benjamin I. P. Rubinstein · 2026-06-08
arXiv:2606. 06833v1 Announce Type: cross Abstract: Automatic Speech Recognition (ASR) systems operating in real-time settings must process acoustic input under strict temporal constraints, where transcription decisions are inherently made on incomplete information.
Read next because Hearing the Unspoken: Language Model Priors for Acoustic Adversarial Attacks overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, under, line, rate, language, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06833v1 Announce Type: cross Abstract: Automatic Speech Recognition (ASR) systems operating in real-time settings must process acoustic input under strict temporal constraints, where transcription decisions are inherently made on incomplete information. This causal constraint serves as an information bottleneck on attackers, significantly limiting attack performance. Our new Semantic Gambit attack breaks this causal limitation by augmenting the adversary with predictive context derived from a Large Language Model in real-time. Our experiments show that this form of augmentation can elevate the corpus-level Word Error Rate to 35.6% -- a three-fold increase over the current state-of-the-art. Ultimately, this work reveals how common, low-latency LLM tooling can be exploited to systematically subvert real-time ASR pipelines.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, adversarial.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07363unread
On the Shoulders of Giants: Empowering Automated Smart Contract Auditing via the GiAnt Corpus
Xiaoting Zhang, Zhipeng Gao, Yiran Lv, Xing Hu, Feifei Niu, Xin Xia · 2026-06-08
arXiv:2606. 07363v1 Announce Type: new Abstract: High-quality smart contract auditing datasets are crucial for evaluating security tools and advancing smart contract security research.
Read next because On the Shoulders of Giants: Empowering Automated Smart Contract Auditing via the GiAnt Corpus overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, rate, extraction, chain. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07363v1 Announce Type: new Abstract: High-quality smart contract auditing datasets are crucial for evaluating security tools and advancing smart contract security research. Two major limitations of existing datasets are the manual-induced scalability bottleneck and the deficiency in data granularity and diversity. To address these limitations, we propose GiANT, an automated framework designed to curate smart contract auditing datasets by distilling vulnerability insights from real-world auditing reports. GiANT employs a divide-and-conquer strategy coupled with the Chain-of-Thought technique to extract structured vulnerability information from Code4rena reports, followed by an LLM-as-a-judge mechanism to perform rigorous quality assurance. To evaluate GiANT's effectiveness, we run it on 388 real-world audit reports and generate the GiAnt Corpus comprising 7,711 vulnerability findings across five severity levels. Manual assessment of the dataset demonstrates exceptional reliability in information extraction, achieving a mean quality score of $4.76\pm0.37$ (out of 5) with inter-rater agreement $\kappa$ of 0.88. We further validate the practicality of our dataset by benchmarking 4 state-of-the-art LLMs on vulnerability detection, code summarization, mitigation recommendation, and automated gas optimization tasks, to establish performance baselines, thereby providing a valuable data foundation for future research in automated smart contract auditing.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations, benchmark.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07341unread
Empirical Evaluation of Large Language Models for Migration of Code Fragments to Post-Quantum Cryptography
Javier Pallar\'es de Bonrostro, Ana I. Gonz\'alez-Tablas, Mar\'ia Isabel Gonz\'alez Vasco · 2026-06-08
arXiv:2606. 07341v1 Announce Type: new Abstract: The transition to post-quantum cryptography (PQC) requires not only replacing vulnerable cryptographic primitives, but also refactoring the surrounding software logic.
Read next because Empirical Evaluation of Large Language Models for Migration of Code Fragments to Post-Quantum Cryptography overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, soft, eval, source, line, rate. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07341v1 Announce Type: new Abstract: The transition to post-quantum cryptography (PQC) requires not only replacing vulnerable cryptographic primitives, but also refactoring the surrounding software logic. While existing PQC migration frameworks provide organizational guidance, practical code-level remediation remains largely manual and error-prone. This paper evaluates whether large language models (LLMs) can be trained to assist in the migration of pre-quantum cryptographic code fragments to post-quantum counterparts while preserving functional correctness. To this end, we introduce a reproducible experimental framework built around a synthetic dataset of 800 paired Python code fragments covering six cryptographic families and combined multi-primitive cases. Each pair is validated through category-specific functional tests, enabling both dataset quality control and objective evaluation of model-generated migrations. Four models are assessed: GPT-4.1 in a zero-shot setting, and fine-tuned versions of GPT-3.5-turbo, GPT-4.1-mini, and CodeLlama-7B-Instruct. The results show that domain-specific fine-tuning is essential for reliable cryptographic migration. The fine-tuned GPT-4.1-mini model achieves the best overall performance, with a mean static similarity of 0.9072 and a dynamic functional correctness rate of 92.5%, substantially outperforming the zero-shot baseline. A complementary validation on six open-source repositories further shows that the approach can produce useful migrations in localized cryptographic modules, while also revealing limitations in larger projects with complex dependencies and cross-module interactions. These findings suggest that fine-tuned LLMs can serve as practical components in future crypto-agile migration pipelines, provided they are coupled with automated verification and dependency-aware validation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations, evaluation.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07335unread
Defending Jailbreak Attacks on Large Language Models via Manifold Trajectory Kinetics
Hangtao Zhang, Yucheng Zhao, Sishun Liu, Ziqi Zhou, Zeyu Ye, Wei Wan, Minghui Li, Shengshan Hu, Yanjun Zhang, Yi Liu, Leo Yu Zhang · 2026-06-08
arXiv:2606. 07335v1 Announce Type: new Abstract: Jailbreak prompts can bypass alignment guardrails in large language models (LLMs) and elicit unsafe outputs, making reliable deployment-time detection critical.
Read next because Defending Jailbreak Attacks on Large Language Models via Manifold Trajectory Kinetics overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, word, under, alignment, line, rate, language. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07335v1 Announce Type: new Abstract: Jailbreak prompts can bypass alignment guardrails in large language models (LLMs) and elicit unsafe outputs, making reliable deployment-time detection critical. Prior detection approaches largely rely on a fixed metric space, e.g., raw inputs, gradients, or hidden features, in which benign and jailbreak prompts are linearly separable. We show this assumption breaks under (i) pseudo-malicious prompts that are benign by intent but contain safety-related keywords, and (ii) adaptive attacks that explicitly optimize against the deployed detector. To overcome this limitation, we shift our focus from identifying a universal metric space to analyzing the more robust neighborhood structure of the underlying data manifold. We present Manifold Trajectory Kinetics (MTK), which treats an LLM as a kinetic system transforming inputs into outputs and detects jailbreaks by tracking how a prompt's neighborhood structure evolves across layers. Benign prompts remain close to benign neighborhoods throughout inference, whereas jailbreak prompts exhibit a characteristic trajectory that begins near malicious seeds and later strategically shifts toward benign neighborhoods to evade refusal.Across four LLMs and ten jailbreak attacks, MTK achieves strong robustness to both failure modes: on pseudo-malicious prompts, it attains a jailbreak true positive rate of 95% at a false positive rate of 5% on benign prompts and 2% on pseudo-malicious prompts, and under adaptive attacks, it maintains a true positive rate of 85%. We further demonstrate the superior performance of MTK for jailbreak detection in vision-language models. Our code is available at https://github.com/Rookie143/mtk.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, limitation, robustness.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07131unread
MalSkillBench: A Runtime-Verified Benchmark of Malicious Agent Skills
Wenbo Guo, Wei Zeng, Chengwei Liu, Xiaojun Jia, Yijia Xu, Lei Tang, Yong Fang, Yang Liu · 2026-06-08
arXiv:2606. 07131v1 Announce Type: new Abstract: AI coding agents such as Claude Code and Gemini CLI increasingly extend themselves with third-party skills: markdown packages bundling natural-language instructions, executable scripts, and tool permissions.
Read next because MalSkillBench: A Runtime-Verified Benchmark of Malicious Agent Skills overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, fires, line, rate, control. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07131v1 Announce Type: new Abstract: AI coding agents such as Claude Code and Gemini CLI increasingly extend themselves with third-party skills: markdown packages bundling natural-language instructions, executable scripts, and tool permissions. Because a skill is at once code and agent-facing instruction, it introduces a supply chain dependency whose risk is neither pure code nor pure prompt. Detection tools have never been measured against verified ground truth spanning this hybrid space, leaving their effectiveness unknown and wild-only evaluations biased. We present MalSkillBench, the first runtime-verified benchmark of malicious agent skills: 3,944 malicious skills labeled along a three-dimensional taxonomy of 108 cells. Of these, 3,214 come from a closed-loop Generate-Verify-Feedback pipeline admitting only samples whose malicious behavior fires inside a Docker sandbox under system-call monitoring and an LLM judge; we add 703 in-the-wild and 4,000 matched benign skills. Our measurements are consistent: code injection reaches 94.5% verification yield but prompt injection only 75.8%, the same fragility that later makes it hard to detect; the wild sample is narrow, dominated by one cryptocurrency-theft campaign (86.6% one behavior, 81% from two accounts) with a small but architecturally new tail attacking the agent control plane; the strongest skill-specific detector reaches 98.4% recall on code injection yet collapses on prompt-injection and agent-control attacks, and wild-only scoring swings the ranking by up to 66 recall points; supply-chain scanners and prompt-injection defenses each see only half of a skill, and no combination recovers the code-instruction relationship. Detecting malicious skills therefore requires reasoning jointly over task intent, code, and instructions. We release the dataset, pipeline, baselines, and results.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias, evaluation, benchmark.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.07009unread
Fast Bounded-Independence Functions and Their Duals
Martijn Brehm, Yuval Ishai, Nicolas Resch · 2026-06-08
arXiv:2606. 07009v1 Announce Type: new Abstract: We continue the study of {\em fast} functions, computable by linear-size circuits, that share useful properties of random functions.
Read next because Fast Bounded-Independence Functions and Their Duals overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, rate. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.07009v1 Announce Type: new Abstract: We continue the study of {\em fast} functions, computable by linear-size circuits, that share useful properties of random functions. Motivated by cryptographic applications, we generalize and improve on previous results in this area, obtaining the following results: - For any constant $t$, we construct a fast $t$-wise independent hash function with algebraic degree $\log_2 t$ (over $\mathbb F_2$), simultaneously optimizing both asymptotic circuit size and degree. - We simplify and improve a recent construction (ITCS 2026) of a family of fast codes with fast duals, both meeting the Gilbert-Varshamov bound. Unlike the previous construction, our construction has negligible failure probability, can accommodate general fields and rates, supports a systematic encoding, and admits fast universal encoders. - We strengthen the above to support stronger random-like properties, such as optimal combinatorial list-decoding. This is achieved by constructing, for any constant $t$, a family of fast linear functions that map any $t$ linearly independent inputs to uniform and statistically independent outputs. Prior to our work, this was only known for $t=1$. We demonstrate the usefulness of the above results to cryptography. This includes the first nontrivial protocols for perfectly secure multiparty computation whose circuit complexity scales linearly with the number of parties, as well as protocols for computing encrypted matrix-vector products with optimal asymptotic circuit complexity.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06968unread
HAVE: Host Active Verification Engine for Closing the Contextual Reality Gap in Security Digital Twins
Vincenzo Sammartino, Marco Pasquini · 2026-06-08
arXiv:2606. 06968v1 Announce Type: new Abstract: Security Digital Twins (SDTs) provide continuously updated virtual replicas of infrastructure for threat simulation, yet they rely on theoretical CVSS scores to assign lateral-movement probabilities -- creating the Contextual Reality Gap: risk is overestimated where unacknowledged mitigations neutralize exploits, and drastically underestimated where logic flaws bypass all memory-safety defenses.
Read next because HAVE: Host Active Verification Engine for Closing the Contextual Reality Gap in Security Digital Twins overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, under, width, correct, alpha, eval. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06968v1 Announce Type: new Abstract: Security Digital Twins (SDTs) provide continuously updated virtual replicas of infrastructure for threat simulation, yet they rely on theoretical CVSS scores to assign lateral-movement probabilities -- creating the Contextual Reality Gap: risk is overestimated where unacknowledged mitigations neutralize exploits, and drastically underestimated where logic flaws bypass all memory-safety defenses. We present the Host Active Verification Engine (HAVE), an SDT extension that deploys a safety-constrained host agent to measure the empirical probability of compromise $\hat{p}$ via maximum-likelihood estimation over snapshot-isolated Bernoulli trials. A Wilson interval-width confidence weight $\alpha_w$ propagates $\hat{p}$ into Monte Carlo simulations via a Bayesian blending rule formally related to the Beta-Binomial posterior. Evaluation across four vulnerability classes, three security tiers, and two production binaries shows HAVE reduces $P_{\text{reach}}$ by 38.2% in false-positive scenarios and increases it by 132.4% in false-negative scenarios, with a net +124.1% correction; post-HAVE estimates vary by only $1.12\times$ across calibration exponents $\kappa$, versus $4.6\times$ for CVSS-only baselines.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses negative, evaluation.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06914unread
DPAgent-in-the-Middle: Agentic Defense and Repair Against AI-Groomed Deceptive Patterns
Zewei Shi, Ruoxi Sun, Haoyang Li, Seong Oun Hwang, Feng Liu, Minhui Xue, Xingliang Yuan · 2026-06-08
arXiv:2606. 06914v1 Announce Type: new Abstract: Privacy deceptive patterns in web interfaces systematically manipulate users into disclosing personal data, yet existing defenses are fragmented, static, and increasingly vulnerable to manipulation by large language models.
Read next because DPAgent-in-the-Middle: Agentic Defense and Repair Against AI-Groomed Deceptive Patterns overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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 "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: persona, rect, eval, middle, line, rate, full, chain. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06914v1 Announce Type: new Abstract: Privacy deceptive patterns in web interfaces systematically manipulate users into disclosing personal data, yet existing defenses are fragmented, static, and increasingly vulnerable to manipulation by large language models. Moreover, data voids, areas of information scarcity within the web ecosystem, create fertile ground for adversaries to inject misleading content that can be scraped and learned by AI systems, thereby amplifying both deceptive design and model misbehavior. In this paper, we formalize a new threat model, AI grooming, where attackers exploit data voids to seed benign-looking but malicious samples that corrupt model reasoning and normalize deceptive practices. To address this threat in privacy deceptive patterns, we present DPAgent, an agentic and reasoning-aware framework that orchestrates four specialized agents to mitigate the AI Grooming threat via a proactive defense that combines latent space purification with defensive prompting and operates directly in live web environments to proactively explore, detect, and repair privacy deceptive user interfaces before they reach end users. Extensive evaluations show that DPAgent detects 90.98% of groomed samples, achieves state-of-the-art privacy deceptive pattern detection with a micro F1 of 0.816, explores over 80% of pattern types while visiting only about 10% of the pages required by baselines, and successfully repairs 77% of detected deceptive interfaces. A large-scale study of 485 websites in the wild reveals that up to 98% contain at least one privacy deceptive pattern, over 90% of which can be mitigated by DPAgent. User studies further confirm that DPAgent effectively reduces privacy risks while preserving browsing experience. Our results demonstrate the promise of agent-in-the-middle defenses for securing the web UI supply chain against deceptive design and emerging AI threats rooted in data void exploitation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06894unread
FDM: A Framework for Decision-making to build ML-based Malware detection systems
Tadiwa Vhito, Jakapan Suaboot, Warodom Werapun, Norrathep Rattanavipanon · 2026-06-08
arXiv:2606. 06894v1 Announce Type: new Abstract: Selecting appropriate machine learning (ML) configurations for malware detection is a complex, multi-criteria problem.
Read next because FDM: A Framework for Decision-making to build ML-based Malware detection systems overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, source, without, contexts, test, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06894v1 Announce Type: new Abstract: Selecting appropriate machine learning (ML) configurations for malware detection is a complex, multi-criteria problem. Model choice, feature engineering, and update mechanisms must jointly satisfy operational constraints that vary across deployment contexts. This paper proposes the Framework for Decision-making (FDM) to build ML-based malware detection systems. The FDM formalises this selection process using the Weighted Configuration Compatibility Score (WCCS), a multi-criteria scoring function mapping five operational parameters (platform constraint, resource budget, response latency, update frequency, and detection sensitivity) to ranked recommendations across nine configuration dimensions. To validate the framework, four experiments were conducted on three datasets (a private Windows API dataset, the public Malimg image benchmark, and an Android static API dataset). Key results include: (i) XGBoost achieved the best accuracy-to-resource ratio in binary classification (97.46 % test accuracy, <70 MB RAM), outperforming LSTM/BiLSTM which consumed up to 2.8 GB; (ii) in multi-class classification, classical models (XGBoost 79.03 %) outperformed recurrent deep models (BiLSTM 72.27 %), reversing the binary ranking; (iii) class-incremental learning with EfficientNetB0 maintained 99.13 % accuracy with only 0.65 pp degradation across 11 incremental steps; (iv) transfer learning reduced training time by 2.14 times on average for image-based malware data without significant accuracy cost; and (v) autoencoder pre-processing yielded a 14 times training speedup at a cost of only 0.86 pp accuracy. These findings confirm that the optimal ML configuration is context-dependent, validating the FDM's core premise and demonstrating its practical utility for cybersecurity practitioners.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06815unread
AMD-FCG: An Enhanced Function Call Graph Dataset with Integrated Topological Features for Malware Detection and Classification
Parthajit Borah, Sakshi Singh, D. K. Bhattacharyya, J. K. Kalita · 2026-06-08
arXiv:2606. 06815v1 Announce Type: new Abstract: As malware illustrates a complex structure and behavior, detection of these has been a significant challenge in the domain of cybersecurity along with related services in daily life.
Read next because AMD-FCG: An Enhanced Function Call Graph Dataset with Integrated Topological Features for Malware Detection and Classification 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, rate. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06815v1 Announce Type: new Abstract: As malware illustrates a complex structure and behavior, detection of these has been a significant challenge in the domain of cybersecurity along with related services in daily life. So, it becomes crucial to have a reliable and adaptive solution to address the issue. Among the several detection methods developed over the years, one of the most reliable ones is studying and analyzing the structural and behavioral patterns of malware. These patterns of sophisticated malware can be obtained with the help of Function Call Graphs (FCGs). However, to effectively cover numerous groups of families of malware, it is required to have a sufficiently large dataset for the system to operate on. In order to ensure accuracy and robustness of the system, the dataset should comprise samples of different malwares and a benign application for secure execution of the detection process. This paper introduces AMD-FCG, an enhanced Function Call Graph dataset integrated with topological features of malwares. The framework enhances the detection procedure, streamlining the workflow for cybersecurity professionals and also eliminating the need for dynamic analysis and extensive processing. Therefore, it can be used to develop and deploy more efficient and innovative malware detection 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.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06784unread
What Your Posts Reveal: A Benchmark and Agentic Framework for User-Level Privacy Leakage on Social Media
Zifan Peng, Yini Huang, Aiwen Lu, Qiming Ye, Peixian Zhang, Jingyi Zheng, Yule Liu, Xuechao Wang, Xinlei He, Jiaheng Wei · 2026-06-08
arXiv:2606. 06784v1 Announce Type: new Abstract: Public social media posts can reveal private information through weak cues scattered across text, images, or metadata.
Read next because What Your Posts Reveal: A Benchmark and Agentic Framework for User-Level Privacy Leakage on Social Media overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, line, leakage. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06784v1 Announce Type: new Abstract: Public social media posts can reveal private information through weak cues scattered across text, images, or metadata. Such leakage is often cumulative and cross-post: cues that appear harmless in isolation may jointly expose a user's home, workplace, or routine. However, current research lacks a unified benchmark for user-level multimodal privacy leakage and an evaluation metric that captures exposure severity beyond binary accuracy. To address these gaps, we propose SopriBench, a synthetic benchmark guided by leakage patterns abstracted from a private reference corpus of Rednote and Instagram accounts, covering 50 user profiles and 1,569 images with attributes, contextual sensitivity, granularity, leakage type, inference difficulty, and supporting evidence. We further introduce the Privacy Exposure Score (PES), which weights value granularity by contextual sensitivity. Inspired by abductive reasoning, we introduce Argus, a training-free agentic framework for cumulative leakage inference. Argus forms hypotheses from accumulated evidence, verifies supporting evidence, and aggregates cross-post cues into privacy profiles, achieving 0.55 PES, a 25% improvement over the strongest baseline, with the largest gain on cross-post leakage.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation, benchmark.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06502unread
Subtle Injection for Ground-truth Inference of LLM Training Data
Abraham Itzhak Weinberg · 2026-06-08
arXiv:2606. 06502v1 Announce Type: new Abstract: As large language models (LLMs) are increasingly trained on scraped web corpora without authorisation, content owners require forensic methods to prove that their documents were included in a model's training set.
Read next because Subtle Injection for Ground-truth Inference of LLM Training Data overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, phrase, under, rate, control, without, trained. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06502v1 Announce Type: new Abstract: As large language models (LLMs) are increasingly trained on scraped web corpora without authorisation, content owners require forensic methods to prove that their documents were included in a model's training set. We propose \textbf{SIGIL} (\textbf{S}ubtle \textbf{I}njection for \textbf{G}round-truth \textbf{I}nference of \textbf{L}LM training data), a framework that embeds imperceptible \emph{canary sequences} into protected text and code such that any LLM trained on those documents exhibits statistically detectable behavioural signatures when probed with targeted queries. SIGIL defines five canary strategies -- lexical-rare, lexical-phrase, syntactic, semantic, and code-pattern -- and a \emph{Membership Inference Score} (MIS) grounded in the Neyman-Pearson hypothesis testing framework with formal false-positive rate (FPR) control. Simulator parameters are calibrated against the empirical membership inference literature, yielding realistic heterogeneous results across $36{,}000$ trials: overall AUC $= 0.892$, rising from $0.831$ at $0.1\%$ injection to $0.947$ at $10\%$. Detection rates range from $33\%$ to $78\%$ across model-size and injection-rate conditions. Code Pattern canaries achieve the highest AUC ($0.903$, Cohen's $d = 1.84$); Syntactic Structure the lowest ($0.875$, $d = 1.63$). All four experimental factors -- injection rate, model size, canary strategy, and mixture ratio -- have significant independent effects on MIS ($p 0.86$ even under $100\%$ paraphrasing ($\text{AUC} = 0.864$), confirming robustness through semantic leakage that survives surface-level rewriting.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses robustness.
- score 100arxiv cs.CR (Cryptography and Security)arxiv:2606.06501unread
Enhancing Malware Detection with Generative AI: Using Variational Autoencoders to Boost Machine Learning Classifiers' Performance
Mohammad Alharbi, Jeremy Straub · 2026-06-08
arXiv:2606. 06501v1 Announce Type: new Abstract: The advancement of malware poses obstacles for cybersecurity, necessitating the development of advanced detection techniques.
Read next because Enhancing Malware Detection with Generative AI: Using Variational Autoencoders to Boost Machine Learning Classifiers' Performance overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, rate, trained, model. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06501v1 Announce Type: new Abstract: The advancement of malware poses obstacles for cybersecurity, necessitating the development of advanced detection techniques. This paper proposes an approach to enhance malware detection through the use of a generative artificial intelligence model. Specifically, variational autoencoders (VAEs) are used with the random forest, XGBoost and sequential model machine learning classifiers. Generated synthetic malware samples are used to address the critical issue of data scarcity for new or less common malware types. This approach can be used to augment datasets to improve classifier robustness. The proposed methodology uses VAEs to create high-quality diverse synthetic datasets that closely mimic real-world malware data. The effectiveness of these augmented datasets is evaluated by comparing the performance of the machine learning classifiers when they are trained with the original data and when they are trained with the synthetic data-augmented datasets. The results demonstrate a notable improvement in the accuracy, precision, recall and F1-scores of the classifiers, when they are trained using the augmented datasets. The enhanced performance for detecting various malware classes shows the potential of this approach to facilitate adaptation to evolving malware threats effectively. This work demonstrates the utility of generative AI for cybersecurity. It also provides a foundation for future research aimed at developing more resilient and adaptive malware detection systems.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses robustness.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07299unread
DuMate-DeepResearch: An Auditable Multi-Agent System with Recursive Search and Rubric-Grounded Reasoning
Lingyong Yan, Can Xu, Yukun Zhao, Wenxuan Li, Qingyang Chen, Jiulong Wu, Wenli Song, Xiangnan Li, Weixian Shi, Yiqun Chen, Xuchen Ma, Yuchen Li, Jiashu Zhao, Shuaiqiang Wang, Jianmin Wu, Dawei Yin · 2026-06-08
arXiv:2606. 07299v1 Announce Type: new Abstract: Deep Research (DR) has emerged as a new agentic paradigm to tackle complex, open-ended research tasks, demanding systems that can iteratively frame problems, acquire evidence, verify sources, and synthesize long-form reports.
Read next because DuMate-DeepResearch: An Auditable Multi-Agent System with Recursive Search and Rubric-Grounded 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: latin, under, eval, source, rate, trained, test. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07299v1 Announce Type: new Abstract: Deep Research (DR) has emerged as a new agentic paradigm to tackle complex, open-ended research tasks, demanding systems that can iteratively frame problems, acquire evidence, verify sources, and synthesize long-form reports. In practice, however, current DR systems are constrained by four interrelated limitations: long-horizon planning over an underspecified scope, the bottleneck of decomposing and scheduling such tasks within a single agent, hallucination risk in long-form synthesis, and limited process auditability. This technical report presents DuMate-DeepResearch, a multi-agent DR framework built on the Qianfan Agent Foundry. The framework decouples the Agent Core, which handles task understanding, planning, and scheduling, from an extensible Tool Ecosystem for retrieval, evidence acquisition, and report rendering, making every intermediate decision and tool invocation explicitly traceable. Building on this infrastructure, DuMate-DeepResearch further introduces three mechanisms: (i) a graph-based dynamic planning strategy expands the research roadmap coarse-to-fine and continuously revises it through reflection, re-planning, backtracking, and parallel branching; (ii) a recursive two-level execution design delegates each complex search sub-task to an inner Search Agent that runs its own planning loop, isolating noisy retrieval and stabilizing long-horizon execution; (iii) a rubric-based test-time optimization mechanism dynamically generates task-specific quality criteria and uses them as live reasoning scaffolds for evidence-grounded synthesis and adaptive stopping. Across two deep research benchmarks, DuMate-DeepResearch establishes new state-of-the-art results: the best overall score (58.03%) on DeepResearch Bench, and the best overall score (61.95%) on DeepResearch Bench II while ranking first in information recall and analysis.
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 limitation, limitations, benchmark.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07253unread
TOPSIS-RAD: Ranking According to Desires
Leonardo Fernandes Costa, Helder Gomes Costa, Diogo Lima, Brunno Rodrigues · 2026-06-08
arXiv:2606. 07253v1 Announce Type: new Abstract: Traditional TOPSIS derives its reference points -- the Positive Ideal Solution ($PIS$) and Negative Ideal Solution ($NIS$) -- from the observed alternative set, making rankings susceptible to misalignment with decision-maker (DM) requirements, sensitivity to outlier performances, and rank reversal.
Read next because TOPSIS-RAD: Ranking According to Desires 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. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07253v1 Announce Type: new Abstract: Traditional TOPSIS derives its reference points -- the Positive Ideal Solution ($PIS$) and Negative Ideal Solution ($NIS$) -- from the observed alternative set, making rankings susceptible to misalignment with decision-maker (DM) requirements, sensitivity to outlier performances, and rank reversal. This paper proposes TOPSIS-RAD, which addresses these issues by incorporating two arrays of DM-defined reference levels. Vetoed Performance Levels ($VPL$) exclude non-viable alternatives before normalisation, preventing them from distorting the ranking frontiers. Desired Performance Levels ($DPL$) cap performances at the DM's desired level before normalisation, anchoring the $PIS$ in explicit aspirations rather than dataset extremes. Three toy examples demonstrate each mechanism: $VPL$ reshapes normalisation boundaries by removing a non-viable alternative; fixed $DPL$ frontiers stabilise rankings by limiting the influence of performances well above the desired level. The method preserves the familiar distance-based structure of TOPSIS while grounding the ranking in stable, DM-specified boundaries. Limitations and future research directions are also discussed.
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, limitations, negative.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07157unread
Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI Models
Dewi Gould, Francis Rhys Ward, Anders Cairns Woodruff, Rauno Arike, Josh Hills, Alex Serrano, Ida Caspary, Jason Ross Brown, Jo J. Jiao, Patrick Leask, Twm Stone, Ram Potham, Ionut Gabriel Stan, Harry Mayne, Simeon Hellsten, Shubhorup Biswas, Ariana Azarbal, William L. Anderson, Elle Najt, Ryan Greenblatt, Julian Stastny · 2026-06-08
arXiv:2606. 07157v1 Announce Type: new Abstract: Many efforts to ensure frontier AI models are safe rely on monitoring their chain-of-thought (CoT) reasoning.
Read next because Think Fast: Estimating No-CoT Task-Completion Time Horizons of Frontier AI 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 "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, compare, project, without, chain, completion. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07157v1 Announce Type: new Abstract: Many efforts to ensure frontier AI models are safe rely on monitoring their chain-of-thought (CoT) reasoning. If models become able to perform sufficiently complex reasoning internally, without explicit thinking tokens, this would undermine such oversight. We measure how well frontier models reason without CoT across a suite of over 30,000 questions spanning 43 benchmarks in domains including math, coding, puzzles, causality, theory-of-mind, and strategic reasoning. To compare models against humans, we estimate the $50\%$-task-completion time horizon (TH): the human time required for tasks a model completes with $50\%$ success rate. We complement this with a $50\%$ reasoning token horizon: the minimum number of o3-mini reasoning tokens needed for tasks a model solves with $50\%$ success rate. We find that the no-CoT $50\%$ TH of frontier models has been doubling roughly every year over the past six years, with GPT-5.5's TH reaching over 3 minutes and reasoning token horizon exceeding 1,500 tokens. Our median estimates predict that frontier no-CoT THs could exceed 7 minutes by 2028, and 25 minutes by 2030, though these projections carry substantial uncertainty. We recommend frontier developers track this explicitly.
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.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07108unread
DyCon: Dynamic Reasoning Control via Evolving Difficulty Modeling
Tengyao Tu, Yulin Li, Hui-Ling Zhen, Libo Qin, Zhoujun Wei, Jinghua Piao, Zhuotao Tian, Yong Li, Min Zhang · 2026-06-08
arXiv:2606. 07108v1 Announce Type: new Abstract: Recent advances in Large Reasoning Models (LRMs) demonstrate remarkable performance improvements by iteratively reflecting, exploring, and executing complex tasks, yet suffer from inefficiencies due to redundant reasoning, known as "overthinking".
Read next because DyCon: Dynamic Reasoning Control via Evolving Difficulty Modeling overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, rate, project, control, without, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07108v1 Announce Type: new Abstract: Recent advances in Large Reasoning Models (LRMs) demonstrate remarkable performance improvements by iteratively reflecting, exploring, and executing complex tasks, yet suffer from inefficiencies due to redundant reasoning, known as "overthinking". Existing methods to mitigate this issue either rely on static difficulty estimates or require task-specific training, and thus fail to adapt to the dynamic complexity during reasoning. In this work, we empirically show that the problem difficulty evolves dynamically throughout the reasoning process and is linearly encoded in the LRM's step-level embeddings. Building on this insight, we propose DyCon, a training-free framework that leverages latent step-level representations to explicitly model the evolving task difficulty, enabling the dynamic control of reasoning depth to mitigate the overthinking issue. Extensive experiments conducted on four models ranging from 4B to 32B, and across twelve benchmarks in math reasoning, general question answering, and coding tasks demonstrate that DyCon significantly enhances reasoning efficiency by reducing redundant steps without sacrificing accuracy or generalization. Project page and code are available at https://github.com/yu-lin-li/DyCon.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07047unread
Front-to-Attractors: Modifying the Front-to-Front Heuristic in Bidirectional Search
Alvin Zou, Muhammad Suhail Saleem, Maxim Likhachev · 2026-06-08
arXiv:2606. 07047v1 Announce Type: new Abstract: Heuristics play a central role in the performance of bidirectional search algorithms, which commonly rely on two main classes.
Read next because Front-to-Attractors: Modifying the Front-to-Front Heuristic in Bidirectional Search 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, compare, full. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07047v1 Announce Type: new Abstract: Heuristics play a central role in the performance of bidirectional search algorithms, which commonly rely on two main classes. Front-to-end (F2E) heuristics estimate the distance from a state s to the target of the search (the goal for forward search or the start for backward search). In contrast, front-to-front (F2F) heuristics estimate the distance from s to the opposite search frontier using a pairwise function h(s, s'), where s' ranges over frontier states. Although F2F heuristics are typically more informative and therefore reduce the number of node expansions, their reliance on extensive pairwise evaluations incurs substantial computational overhead. To address this limitation, we introduce a new heuristic class, front-to-attractors (F2A), that preserves much of the informativeness of F2F while dramatically reducing its computational cost. Rather than evaluating distances to all states on the opposite frontier, F2A estimates the distance from s to a small, dynamically maintained set of attractors in the opposite search direction. These attractors serve as a surrogate for the full frontier, enabling rich heuristic guidance at a fraction of the computational expense while maintaining the optimality guarantees offered by F2F. We evaluate F2A across multiple domains and show that it reduces the number of pairwise evaluations by up to 11.2x compared to F2F, while achieving 4.8x fewer node expansions than F2E on average.
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 limitation, evaluation.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07033unread
Hierarchical Semantic-Constrained Heterogeneous Graph for Audio-Visual Event Localization
Zhe Yang, Ruyi Zhang, Hongtao Chen, Wenrui Li, Hengyu Man, Wangmeng Zuo, Xiaopeng Fan · 2026-06-08
arXiv:2606. 07033v1 Announce Type: new Abstract: Open-vocabulary audio-visual event localization (OV-AVEL) jointly models audio-visual cues to recognize and temporally localize events, including categories unseen during training.
Read next because Hierarchical Semantic-Constrained Heterogeneous Graph for Audio-Visual Event Localization overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, rate, trained, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07033v1 Announce Type: new Abstract: Open-vocabulary audio-visual event localization (OV-AVEL) jointly models audio-visual cues to recognize and temporally localize events, including categories unseen during training. Existing methods primarily learn joint audio-visual representations in Euclidean space, but still face two significant challenges. First, the lack of supervision signals for unseen categories makes it difficult to maintain audio-visual consistency across multiple temporal scales. Second, the lack of hierarchical constraints between segment- and video-level semantics prevents the model from establishing semantic consistency across different levels. To address these challenges, we propose a hierarchical semantic constrained heterogeneous graph (HSCHG) for audio-visual event localization framework. We first construct a heterogeneous hierarchical graph in Euclidean space, which includes audio and visual segment nodes and their corresponding video-level nodes. We use multi-directional temporal edges to capture complete temporal information within each modality. Simultaneously, we employ a dual-threshold filtering gated fusion strategy, introducing cross-modal information only when the alignment confidence is high. Furthermore, we introduce bidirectional semantic constraints between segment- and video-level representations to achieve semantic consistency across different levels. Based on this, we map the multi-level audio-visual representations and text prototypes uniformly into hyperbolic space. We use a hierarchical entailment regularization loss to characterize the hierarchical relationships between videos and segments. Extensive experimental results show that our method outperforms existing methods on the OV-AVEL benchmark. Ablation studies further validate the effectiveness of our method.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07027unread
StainFlow: Entity-Stain Tracking and Evidence Linking for Process Rewards in GUI Agents
Haojie Hao, Longkun Hao, Yihang Lou, Yan Bai, Zhenyang Li, Zhichao Yang, Dongshuo Huang, Hongyu Lin, Lanqing Hong, Jiakai Wang, Xianglong Liu · 2026-06-08
arXiv:2606. 07027v1 Announce Type: new Abstract: Reinforcement Learning (RL) has become a promising approach for improving GUI Agents in long-horizon, stochastic digital environments, but trajectory-level success feedback is too sparse to provide reliable credit assignment for intermediate exploration steps.
Read next because StainFlow: Entity-Stain Tracking and Evidence Linking for Process Rewards in GUI Agents 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 "#351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation". Matching terms: eval, line, rate, position, candidate, lora, completion, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07027v1 Announce Type: new Abstract: Reinforcement Learning (RL) has become a promising approach for improving GUI Agents in long-horizon, stochastic digital environments, but trajectory-level success feedback is too sparse to provide reliable credit assignment for intermediate exploration steps. To mitigate this issue, recent studies introduce Process Reward Models (PRMs), which provide finer-grained training feedback through global milestone verification or local step-level evaluation. However, these methods still suffer from two level-specific limitations: global milestone decomposition is subjective and singular, making it difficult to accommodate the multiple valid execution paths in real GUI tasks, while fixed local judging windows may miss long-range key evidence or dilute the decision signal with irrelevant frames. Inspired by stain-tracing mechanisms in network flow analysis, we propose StainFlow, an entity-stain-flow process reward model for GUI Agents. To reduce the subjectivity of global partitioning, we introduce the Global Entity Stain Tracking module, which extracts visually verifiable task entities and tracks how their stain concentrations and states evolve along the trajectory, allowing task phases to be objectively separated by changes in the entity evidence flow. To improve the accuracy of local verification, we introduce the Local Stain Evidence Linking module. Centered on the triggering entities of each candidate key node, it retrieves relevant steps based on their stain concentrations and state changes, and dynamically constructs high-density evidence windows for verifying true key nodes. Extensive experiments on AndroidWorld and OGRBench show that StainFlow relatively improves online RL success by 3.2% and trajectory completion judgment accuracy by 1.8%.
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 limitation, limitations, evaluation.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07017unread
The Sim-to-Real Gap of Foundation Model Agents: A Unified MDP Perspective
Xiaoou Liu, Tiejin Chen, Weibo Li, Xiyang Hu, Hua Wei · 2026-06-08
arXiv:2606. 07017v1 Announce Type: new Abstract: Foundation model agents are increasingly deployed for real-world decision-making, but suffer from the sim-to-real gap.
Read next because The Sim-to-Real Gap of Foundation Model Agents: A Unified MDP Perspective 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, eval, rate, control, test, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07017v1 Announce Type: new Abstract: Foundation model agents are increasingly deployed for real-world decision-making, but suffer from the sim-to-real gap. While robotics and classical control have mature frameworks to address this gap, the foundation model community is treating agent robustness as an entirely novel phenomenon. Our paper proposes formalizing the foundation model agent evaluation and training gap as a classical sim-to-real problem structured entirely around the four elements of a Markov Decision Process, including Observation, Action, Transition, and Reward. In this paper, we set a comprehensive research agenda that translates classical discrepancies into the foundation model domain and advocates for adopting established solutions like domain randomization. We provide concrete examples, such as a multilingual tool calling to demonstrate how severe observation space gaps lead to operationally invalid actions despite correct semantic intent. Ultimately, this agenda aims to drive a paradigm shift, yielding a unified vocabulary and standardized stress test benchmarks to foster a new generation of highly trustworthy agents for reliable real-world applications.
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, evaluation, benchmark.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.07000unread
Teaching the Way, Not the Answer: Privileged Tutoring Distillation for Multimodal Policy Optimization
Shizhe Xiang, Ke An, Wenlong Yu, Yue Liu, Jian Luan, Pei Fu, Qilong Wang · 2026-06-08
arXiv:2606. 07000v1 Announce Type: new Abstract: Recent post-training methods, particularly Reinforcement Learning with Verifiable Rewards (RLVR), have significantly enhanced the reasoning ability of Large Vision-Language Models (LVLMs).
Read next because Teaching the Way, Not the Answer: Privileged Tutoring Distillation for Multimodal Policy Optimization overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, alignment, token, line, without, contexts, lora. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.07000v1 Announce Type: new Abstract: Recent post-training methods, particularly Reinforcement Learning with Verifiable Rewards (RLVR), have significantly enhanced the reasoning ability of Large Vision-Language Models (LVLMs). However, the sparse nature of verifiable rewards provides little token-level supervision for failed rollouts, often leading to inefficient exploration in complex multimodal reasoning tasks. Although policy distillation can offer dense guidance, external teacher based methods introduce substantial computational overhead, while answer conditioned tuning methods may expose answer-level information and induce shortcut-like generation behavior. To address these limitations, we propose PTD-PO, a Privileged Tutoring Distillation Policy Optimization framework for RLVR that provides dense guidance without exposing the answer to the student policy. Specifically, PTD-PO constructs structured privileged hints from spatial attention guidance and intermediate textual reasoning steps, and uses them through in-context learning to produce step-wise token-distribution supervision. The student is still optimized under the original answer-free context, and its failed rollouts are aligned with the hint-augmented reference model at the token-distribution level. To further stabilize distillation under the distribution shift between guided and unguided contexts, we introduce a Top-K Jensen-Shannon divergence objective that focuses alignment on informative token probabilities while reducing memory overhead. Experiments on LVLMs ranging from 2B to 8B parameters show that PTD-PO consistently outperforms RLVR and distillation baselines, mitigates entropy collapse, and improves complex multimodal reasoning performance.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06976unread
Exploring Agentic Tool-Calling Decisions via Uncertainty-Aligned Reinforcement Learning
Yijin Zhou, Linqian Zeng, Xiaoya Lu, Wenyuan Xie, Dongrui Liu, Junchi Yan, Jing Shao · 2026-06-08
arXiv:2606. 06976v1 Announce Type: new Abstract: Large language model (LLM)-based agents often make suboptimal tool-use decisions, including unsupported tool invocation and hallucinated direct responses, which may accumulate errors throughout multi-step interactions.
Read next because Exploring Agentic Tool-Calling Decisions via Uncertainty-Aligned Reinforcement 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 "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, rate, lora, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06976v1 Announce Type: new Abstract: Large language model (LLM)-based agents often make suboptimal tool-use decisions, including unsupported tool invocation and hallucinated direct responses, which may accumulate errors throughout multi-step interactions. Existing approaches mainly improve these behaviors through inference-time correction or coarse-grained reward signals based on decision outcomes and structured checklists, leaving the uncertainty characteristics of agent decisions underexplored. We observe that decision-oriented reinforcement learning tends to weaken the uncertainty separation between correct and incorrect actions, resulting in overconfident mistakes and weaker exploration signals. Therefore, we propose TRUST, which incorporates uncertainty quantification into reward design as a repulsive force for maintaining uncertainty separation, and labels lightweight key-turn annotations for unified post-training of multi-turn trajectories. Experimental results across diverse tool-use benchmarks show that TRUST consistently enhances both decision quality and agent performance while maintaining more reliable uncertainty estimates during optimization.
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.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06972unread
Accounting for Context: Shaping Moral Credences for Value Alignment
Jazon Szabo, Sanjay Modgil · 2026-06-08
arXiv:2606. 06972v1 Announce Type: new Abstract: Ensuring that agent behaviours are aligned with human moral values inevitably raises the problem of how to account for the plurality of moral perspectives that societies -- and even individuals -- typically adopt.
Read next because Accounting for Context: Shaping Moral Credences for Value Alignment overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, alignment, eval, rate, does, factor. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06972v1 Announce Type: new Abstract: Ensuring that agent behaviours are aligned with human moral values inevitably raises the problem of how to account for the plurality of moral perspectives that societies -- and even individuals -- typically adopt. Work on moral uncertainty proposes mechanisms to fairly and democratically aggregate evaluations of actions across different moral theories. However, this paper argues that one needs to account for contextual factors when aggregating moral evaluations. For example, consequentialist perspectives assume an ability to accurately determine how an agent's actions change the world; an assumption that often does not hold in real world settings. We, therefore, formalise agent decision making under moral uncertainty, while also accounting for these kinds of contextual factors. We thereby show that a seemingly commonsensical property -- the weak Pareto principle -- is violated. We argue that this apparent problem is, in fact, a variation of Simpson's paradox, and hence reveals the limitations of aggregation mechanisms that ignore the impact of contextual factors.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations, evaluation.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06941unread
Quantum-Inspired Trace-Augmented Evidence Selection for Reasoning over Structured Hypothesis Spaces
Laura Wynter, Nirvik Sahoo, Paul Griffin · 2026-06-08
arXiv:2606. 06941v1 Announce Type: new Abstract: Large language models (LLMs) now solve a wide range of expert-level exams at or above human level, yet remain brittle on specialised, evidence-intensive domains such as law.
Read next because Quantum-Inspired Trace-Augmented Evidence Selection for Reasoning over Structured Hypothesis Spaces overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, rate, chain, trained. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06941v1 Announce Type: new Abstract: Large language models (LLMs) now solve a wide range of expert-level exams at or above human level, yet remain brittle on specialised, evidence-intensive domains such as law. On these tasks, errors arise not only from gaps in world knowledge but also from subtle distinctions between pieces of evidence and inconsistent use of supporting evidence. The most common aggregator over sampled chain-of-thought (CoT) traces, majority vote, returns the most popular answer regardless of whether its evidence is actually strongest. We propose to treat the selection of CoT reasoning fragments into a set of evidence as an explicit combinatorial optimisation problem, allowing well-supported but minority hypotheses to override noisy majorities, and to evaluate the approach on legal-reasoning benchmarks that are particularly sensitive to evidence quality. We introduce EP-HUBO (Evidence Pool Higher-Order Binary Optimisation), which generates multiple CoT traces with a small local model, parses fragments into per-hypothesis evidence pools, solves a higher-order unconstrained binary optimisation per pool with quality-derived weights (relevance, specificity, distinctiveness), and delegates a single adjudication call per question to a frontier model. We evaluate EP-HUBO on two evidence-intensive legal benchmarks using both simulated annealing on classical hardware and the Dirac-3 photonic entropy-quantum machine from Quantum Computing Inc. HUBO-style optimisation gives a principled way to aggregate reasoning fragments while preserving minority-but-correct hypotheses, and is most valuable in low-contamination domains where frontier models have not already absorbed the benchmark material.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06923unread
Declarative Skills for AI Agents in Knowledge-Grounded Tool-Use Workflows
M. Danish Lim, I. Danial Bin Sharudin, Wen Han Chen, Cedric Lim, Laura Wynter · 2026-06-08
arXiv:2606. 06923v1 Announce Type: new Abstract: We study orchestration mechanisms for tool-using AI agents in realistic customer-service workflows over an unstructured knowledge base.
Read next because Declarative Skills for AI Agents in Knowledge-Grounded Tool-Use Workflows 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, under, eval, line, compare, control, does, test. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06923v1 Announce Type: new Abstract: We study orchestration mechanisms for tool-using AI agents in realistic customer-service workflows over an unstructured knowledge base. We argue that declarative agents -- AI agents equipped with natural-language skill files appended to the system prompt -- are an effective orchestration paradigm. Concretely, we compare (i) a DeclarativeAgent that reads three domain-specific skill files at inference time and decides its own control flow, (ii) an ImperativeAgent based on a programmatic state machine with explicit phases, and (iii) an unscaffolded baseline agent modeled after the $\tau$-Knowledge benchmark agent. Our ImperativeAgent is motivated by externalised-control inference as in Recursive Language Models and graph-based orchestration frameworks. We formalise the three agents as policy classes within a decentralised partially-observable Markov decision process and analyse their information-theoretic and structural properties; we then test the predicted differences empirically on five language models and two retrieval regimes. Our results show that retrieval quality is a dominant bottleneck for AI agents: when evidence is incomplete or skewed, all agents degrade substantially, and skill files cannot recover lost performance. Under high-quality retrieval, however, declarative skills consistently improve accuracy on procedural tasks and reduce orchestration errors, while the imperative state machine's brittleness does not reliably improve task success or compliance.
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.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06869unread
Evidence-Based Intelligent Diagnostic and Therapeutic Visualization System with Large Language Models: Multi-Turn Interaction and Multimodal Treatment Plan Generation
Yunhan Wang, Yuda Wang, Zhiying Tu, Mingqiang Song, Li Song, Kun Li, Dianhui Chu, Bolin Zhang · 2026-06-08
arXiv:2606. 06869v1 Announce Type: new Abstract: Aim: Existing AI-assisted traditional Chinese medicine diagnostic tools suffer from opaque reasoning processes, passive interaction, and limited treatment plan presentation.
Read next because Evidence-Based Intelligent Diagnostic and Therapeutic Visualization System with Large Language Models: Multi-Turn Interaction and Multimodal Treatment Plan Generation 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 "#351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation". Matching terms: eval, line, rate, stage, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06869v1 Announce Type: new Abstract: Aim: Existing AI-assisted traditional Chinese medicine diagnostic tools suffer from opaque reasoning processes, passive interaction, and limited treatment plan presentation. This study proposes a knowledge-enhanced visual diagnostic system to improve the transparency and interpretability of syndrome differentiation and treatment. Methods: The system is built upon a Neo4j knowledge graph comprising 241 syndromes, 1,263 symptoms, and 2,485 relations. It incorporates a four-stage symptom matching pipeline (exact, semantic, fuzzy, and large language model verification), an information gain-driven proactive questioning strategy optimized with genetic algorithms, and a multimodal treatment presentation integrating artificial intelligence-generated illustrations, three-dimensional meridian-acupoint models, and evidence-based literature. Results: Knowledge graph constraints reduced non-standard outputs by 32%. Case studies validated the effectiveness of the interactive workflow across patient self-assessment, clinician-assisted diagnosis, and traditional Chinese medicine education. Automated paired-comparison evaluation across 30 cases further demonstrated significant improvements in diagnostic trust (Cohen's d = 1.82, p < 0.001), reduced cognitive load (improvements in four of five dimensions), and higher credibility of evidence-based references (4.21 vs. 2.95). Conclusions: The proposed system enhances the transparency of traditional Chinese medicine diagnostic reasoning and the interpretability of treatment plans through knowledge graph-driven visualization and multimodal interaction, offering a practical solution for trustworthy artificial intelligence-assisted traditional Chinese medicine applications.
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 evaluation.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06787unread
AdMem: Advanced Memory for Task-solving Agents
Runzhe Wang, Huilin Lu, Shengjie Liu, Li Dong, Jason Zhu · 2026-06-08
arXiv:2606. 06787v1 Announce Type: new Abstract: Large Language Models (LLMs) show promise as tool-using agents but remain limited in long-horizon tasks that require remembering, organizing, and reusing knowledge.
Read next because AdMem: Advanced Memory for Task-solving Agents 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: eval, line, rate, compare, without, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06787v1 Announce Type: new Abstract: Large Language Models (LLMs) show promise as tool-using agents but remain limited in long-horizon tasks that require remembering, organizing, and reusing knowledge. Prior memory approaches aim to resolve the situation, but mainly focus on storing factual information. Recent work on procedural memory improves task reuse, yet often reduces to replaying past successes without addressing failure cases or online scalability. We introduce a unified and automatic memory framework that integrates semantic, episodic, and procedural memory in a bi-level design combining short-term and long-term stores. A multi-agent architecture with actor, memory, and critic agents enables automatic memory generation, reward annotation, and adaptive retrieval. Long-term memory is managed through reward-based evaluation, merging, and pruning, ensuring scalability and continual improvement. Experiments across various environments show that our approach improves robustness and success on long multi-turn tasks compared to existing baselines. This work highlights the importance of comprehensive, adaptive memory for advancing LLM-based agents.
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 failure, robustness, evaluation.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06741unread
OpenSkill: Open-World Self-Evolution for LLM Agents
Zhiling Yan, Dingjie Song, Hanrong Zhang, Wei Liang, Yuxuan Zhang, Yutong Dai, Lifang He, Philip S. Yu, Ran Xu, Xiang Li, Lichao Sun · 2026-06-08
arXiv:2606. 06741v1 Announce Type: new Abstract: Self-evolving agents requires adaptation after deployment, but existing approaches assume a usable learning loop, such as curated skills, successful trajectories, or verifier signals.
Read next because OpenSkill: Open-World Self-Evolution for LLM Agents 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, without, model, never. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06741v1 Announce Type: new Abstract: Self-evolving agents requires adaptation after deployment, but existing approaches assume a usable learning loop, such as curated skills, successful trajectories, or verifier signals. Real open-world deployments may provide none of these, offering only a task prompt. In this work, we study open-world self-evolution, where an agent must build both its skills and its own verification signals from scratch, using open-world resources but no target-task supervision. We propose OpenSkill, a framework that bootstraps this loop: it acquires grounded knowledge and verification anchors from documentation, repositories, and the web, synthesizes them into transferable skills, and refines those skills against self-built virtual tasks grounded in the anchors rather than in target answers. The open world thus supplies both the knowledge to be learned and a supervision-independent practice environment, with target-task supervision reserved for final evaluation. Across three benchmarks and two target agents, OpenSkill attains the best automated pass rate while satisfying the no-supervision constraint. Analysis shows its skills transfer across models without model-specific adaptation, and its self-built verifier aligns with ground-truth outcomes despite never accessing them.
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 evaluation, benchmark.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06735unread
A Geometric Account of Activation Steering through Angle-Norm Decomposition
Georgii Aparin, Tatiana Gaintseva · 2026-06-08
arXiv:2606. 06735v1 Announce Type: new Abstract: Linear activation steering has gained popularity as a simple and empirically effective way to control language model behavior.
Read next because A Geometric Account of Activation Steering through Angle-Norm Decomposition 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, token, line, control, does, position, language. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06735v1 Announce Type: new Abstract: Linear activation steering has gained popularity as a simple and empirically effective way to control language model behavior. More recently, spherical steering paradigms have been proposed to address limitations of additive interventions, often motivated by the assumption that hidden-state norm does not carry concept-relevant information. In this work, we revisit this assumption through a controlled empirical study designed to disentangle the roles of angular and radial components. We show that steering methods differ mainly in how they couple two geometric effects: changing a token's angular alignment with a concept direction and changing its hidden-state norm. Across seven language models, we find that concepts are represented primarily in angular structure, supporting the motivation for spherical methods, but that norm remains important for the stability and downstream effects of steering. Our results explain why interventions with similar concept-level effects can behave differently, and suggest that activation steering should be parameterized by interpretable angular and radial components of the intervention, rather than by a single additive coefficient that entangles these two effects.
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, limitations.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06660unread
AEGIS: A Backup Reflex for Physical AI
Josef Chen · 2026-06-08
arXiv:2606. 06660v1 Announce Type: new Abstract: Long-horizon robot manipulation tends to fail gradually: one bad step degrades the state, and the policy spirals into a basin from which it cannot recover.
Read next because AEGIS: A Backup Reflex for Physical AI overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, control, alone, full, test. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06660v1 Announce Type: new Abstract: Long-horizon robot manipulation tends to fail gradually: one bad step degrades the state, and the policy spirals into a basin from which it cannot recover. The failure is often visible before it happens. We introduce AEGIS (Activation-probe Early-warning, Gated Inference Switching), a selective escalation method that uses a lightweight probe on a weak policy's frozen activations to detect high-risk steps while there is still time to act. When the probe flags a step, control switches to a stronger separate policy, but only for the steps that need it. On LIBERO-Spatial, AEGIS recovers 10.1% of the trajectories the weak policy alone loses, versus 4.6% for budget-matched blind escalation and 5.1% for a random-trigger placebo. These gains are significant under one-sided exact paired McNemar tests with Holm-Bonferroni adjustment over three pre-registered contrasts: +5.4pp over blind escalation, p=8.5e-6; +5.0pp over random triggering, p=1.0e-4; paired-trajectory bootstrap CIs exclude zero. AEGIS activates the stronger policy on only 38% of steps, so the lever is timing rather than compute. The probe clears its precondition with an early-window AUROC of 0.764, 95% CI [0.70, 0.84], read from the weak-policy path over the first 30% of trajectory steps before any handoff. We pre-register the full analysis plan, including a conditional recovered-task-rate estimand and explicit kill criteria, and confirm the result on 700 common-random-number episodes per arm, with nA-fail=646.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06641unread
Accelerated Fourier SAT (AFSAT): Fully Realising a GPU-based Symmetric Pseudo-Boolean SAT Solver
Cody J Christopher, Charles Gretton · 2026-06-08
arXiv:2606. 06641v1 Announce Type: new Abstract: We present Accelerated Fourier SAT (AFSAT), a GPU-accelerated solver for pseudo-Boolean satisfiability based on continuous local search (CLS).
Read next because Accelerated Fourier SAT (AFSAT): Fully Realising a GPU-based Symmetric Pseudo-Boolean SAT Solver 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 "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?". Matching terms: line, rate, implement, full, length, position, candidate. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06641v1 Announce Type: new Abstract: We present Accelerated Fourier SAT (AFSAT), a GPU-accelerated solver for pseudo-Boolean satisfiability based on continuous local search (CLS). AFSAT realises the proof-of-concept approach, FastFourierSAT, into a fully-engineered solver supporting any heterogeneous mixture of symmetric constraint types and lengths within a single problem instance. Using the JAX compiler, AFSAT leverages pure function composition, automatic vectorisation, automatic differentiation, and just-in-time (JIT) compilation to perform massively parallel CLS across batches of candidate assignments. We demonstrate substantially improved numerical stability, runtime performance, and memory efficiency over the proof-of-concept. We achieve this by way of identifying and addressing various limitations that arise from memory latency and floating-point representation, as well as leveraging automatic parallelisation and compact representations. The inherent representational and stability limitations of floating point are partially addressed by a tailored discrete Fourier transform implementation. We achieve near-linear throughput when scaling to multiple accelerators via JAX array sharding.
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 limitation, limitations.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06533unread
Position: Don't Just "Fix it in Post": A Science of AI Must Study Training Dynamics
Stella Biderman, Mohammad Aflah Khan, Niloofar Mireshghallah, Catherine Arnett, Fazl Barez, Naomi Saphra · 2026-06-08
arXiv:2606. 06533v1 Announce Type: new Abstract: What would it mean to have a scientific understanding of AI?
Read next because Position: Don't Just "Fix it in Post": A Science of AI Must Study Training Dynamics overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, wrong, position, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06533v1 Announce Type: new Abstract: What would it mean to have a scientific understanding of AI? Models are not static objects: they are snapshots of time-evolving processes shaped by data, objectives, architectures, and optimization dynamics. Yet much of AI research treats models as fixed artifacts, analyzing behaviors after training rather than asking why they emerge. This position paper argues that a science of AI must move beyond post-hoc fixes and study the training dynamics that produce model behavior. Such a science should support progressively stronger forms of understanding: predicting outcomes from early training signals, intervening when trajectories go wrong, and ultimately designing training procedures that more reliably produce desired properties. Scaling laws have made prediction routine for loss; the challenge is extending this success to capabilities, biases, robustness, and safety-relevant behaviors. We articulate requirements for such theories grounded in the history and philosophy of science, examine progress in mechanistic interpretability, fairness, memorization, and simplicity bias, and identify concrete open problems.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias, robustness.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06529unread
Attack Selection in Agentic AI Control Evaluations Meaningfully Decreases Safety
Catherine Ge-Wang, Tyler Crosse, Benjamin Hadad IV, Joachim Schaeffer, Ram Potham, Tyler Tracy · 2026-06-08
arXiv:2606. 06529v1 Announce Type: new Abstract: An attacker that strategically chooses when to attack is much harder to catch than one that attacks indiscriminately.
Read next because Attack Selection in Agentic AI Control Evaluations Meaningfully Decreases Safety 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, control, without, full, capability, test. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06529v1 Announce Type: new Abstract: An attacker that strategically chooses when to attack is much harder to catch than one that attacks indiscriminately. AI control is a safety framework for deploying capable but untrusted AI agents under the oversight of a weaker, trusted monitor and a limited human audit budget. Control evaluations stress-test these protocols by pitting a red-team attack policy against the blue-team monitor, but current evaluations typically assume attackers that do not strategically select when to attack. We study this capability, attack selection, in agentic settings by decomposing attack decisions into a start policy, which decides when an attacker should attack, and a stop policy, which decides when an attacker should abort an ongoing attack. Across two agentic settings, BashArena and LinuxArena, both policies substantially lower measured empirical safety without changing the underlying attack capability. At a 1% audit budget, our start policy reduces safety by 20pp on both BashArena and LinuxArena, and our stop policy reduces safety by 20pp on BashArena and 28pp on LinuxArena. These reductions should be interpreted as upper bounds on the effect of attack selection. Existing control evaluations may therefore yield overly optimistic safety estimates against selective attackers. We recommend that future evaluations, system cards, and safety cases elicit attack selection to produce more realistic safety estimates.
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.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06526unread
CrowdMath: A Dataset of Crowdsourced Mathematical Research Discussions
Sherin Muckatira, Jesse Geneson, Slava Gerovitch, Pavel Etingof, Mikhail Gronas, Anna Rumshisky · 2026-06-08
arXiv:2606. 06526v1 Announce Type: new Abstract: Large language models have made substantial progress on mathematical reasoning, but existing benchmarks typically evaluate well-specified problems with final answers, step-by-step solutions, or complete proofs.
Read next because CrowdMath: A Dataset of Crowdsourced Mathematical Research Discussions 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, under, eval, source, chain, completion, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06526v1 Announce Type: new Abstract: Large language models have made substantial progress on mathematical reasoning, but existing benchmarks typically evaluate well-specified problems with final answers, step-by-step solutions, or complete proofs. They do not capture collaborative open-problem solving: a setting in which participants propose partial arguments, identify gaps or errors in prior steps, repair flawed reasoning, and gradually synthesize incremental contributions into a proof. We introduce CrowdMath, a dataset of 164 expert-annotated progress chains from the MIT PRIMES--Art of Problem Solving (AoPS) CrowdMath program (2016-2025), a collaborative research initiative whose discussions have led to peer-reviewed publications. Each chain traces a multi-participant forum discussion from an open-problem statement to a completed proof. Posts are labeled by their functional roles in the evolving solution process, including partial progress, proof completion, erroneous reasoning, and error identification. We define evaluation tasks and benchmark six frontier models. Models achieve 83-88% accuracy on next-post prediction, suggesting that they can follow the local flow of mathematical discussion. However, they struggle to identify the functional significance of individual contributions with the best model achieving only 0.42 macro-F1 on post-role classification. CrowdMath exposes a gap between solving well-specified mathematical problems and understanding collaborative mathematical progress as it unfolds.
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.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06523unread
Lean4Agent: Formal Modeling and Verification for Agent Workflow and Trajectory
Ruida Wang, Jerry Huang, Pengcheng Wang, Xuanqing Liu, Luyang Kong, Tong Zhang · 2026-06-08
arXiv:2606. 06523v1 Announce Type: new Abstract: Equipping Large Language Models (LLMs) to execute reliable multi-step workflows has become a central challenge in artificial intelligence.
Read next because Lean4Agent: Formal Modeling and Verification for Agent Workflow and Trajectory 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 "Can capability be taught through another persona?", 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, capability, language, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06523v1 Announce Type: new Abstract: Equipping Large Language Models (LLMs) to execute reliable multi-step workflows has become a central challenge in artificial intelligence. Despite recent advances in LLMs' agentic capabilities, most agent systems still lack formal methods for specifying, verifying, and debugging their workflow and execution trajectories. This challenge mirrors a long-standing problem in mathematics, where the ambiguity of natural languages (NLs) motivates the development of formal languages (FLs). Inspired by this paradigm, we propose **Lean4Agent**, to the best of our knowledge, the first framework that uses Lean4, a dependent-type FL to model and verify agent behavior. **Lean4Agent** launches **FormalAgentLib**, an extensible Lean4 library for formally modeling and verifying agent workflows' semantic consistency under explicit assumptions, and enabling localization of execution-time failures revealed by trajectories. Building on **FormalAgentLib**, we further develop **LeanEvolve**, which applies results in **FormalAgentLib** to revise workflows to enhance its capability. Extensive experiments on a hard problem subset of SWE-Bench-Verified and a subset of ELAIP-Bench across 5 leading LLMs indicate that the verification-passing workflows outperform the failing ones by an average of **11.94%**, and **LeanEvolve** further improves SWE performance by **7.47%** on average. Furthermore, **Lean4Agent** establishes a foundation for a new field of using expressive dependent-type FL to formally model and verify agent behavior.
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.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06518unread
DiBS: Diffusion-Informed Branch Selection
Bo Liu, Yuan Xie, Yuan Gao, Xiaolong Luo, Peng Ye, Tao Chen, Fujun Han · 2026-06-08
arXiv:2606. 06518v1 Announce Type: new Abstract: Sudoku is a representative constraint satisfaction problem that requires global structural reasoning under strict discrete constraints.
Read next because DiBS: Diffusion-Informed Branch Selection overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, under, correct, line, candidate, model. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06518v1 Announce Type: new Abstract: Sudoku is a representative constraint satisfaction problem that requires global structural reasoning under strict discrete constraints. The existing works of solving Sudoku mainly focus on two dominant approaches, i.e., traditional heuristic and deep learning solver. However, they suffer from two complementary limitations: learning-based solvers lack hard correctness guarantees, while complete symbolic solvers are still prone to long-tail search. To address these shortcomings, we propose a novel diffusion model-guided approach, termed as DiBS, for the branch selection search process. Specifically, DiBS keeps the symbolic solver complete and uses the diffusion model as a branch-ordering guide. The core method is ranking candidate values under the current partial assignment and lightweight consistency signal. Furthermore, we provide an in-depth theoretical proof to reveal how it works and why it works. Experiments on the challenging Royle 17-clue Sudoku benchmark show that our DiBS substantially reduces search cost relative to strong heuristic baselines, especially in nodes, backtracks, and long-tail percentiles. Besides, these results confirm that learned global guidance is effective on hard instances where branch-order mistakes are most expensive. All codes are available at https://github.com/shanxierdan/DiBS.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations, benchmark.
- score 100arxiv cs.AI (Artificial Intelligence)arxiv:2606.06514unread
Detecting and Mitigating Bias by Treating Fairness as a Symmetry Operation
Nishit Singh · 2026-06-08
arXiv:2606. 06514v1 Announce Type: new Abstract: Machine learning systems deployed in high stakes socioeconomic settings routinely display bias.
Read next because Detecting and Mitigating Bias by Treating Fairness as a Symmetry Operation overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, source, implement, does, contexts. Source: arxiv cs.AI (Artificial Intelligence).
arXiv:2606.06514v1 Announce Type: new Abstract: Machine learning systems deployed in high stakes socioeconomic settings routinely display bias. We formalize bias as a symmetry breaking operation: a classifier is fair if its outputs remain invariant under the counterfactual operation of switching a sensitive attribute, with merit features held fixed. We implement loss based regularization as a symmetry restoring mechanism and evaluate the framework on four synthetic datasets with varying levels of noise, correlation, and bias. The framework achieves upwards of 90\% violation reduction, with accuracy costs around 5\%. This framework does not require causal graph knowledge, is computationally lightweight, and generalizes to any sensitive attribute definable as a bit-flip, making it suitable for contexts where local sources of discrimination remain absent from mainstream benchmarks.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias, benchmark.
- score 100arxiv cs.CL (NLP)arxiv:2606.06879unread
An Expanded Synthetic Conversation Dataset for Multi-Turn Smishing Detection
Carl Lochstampfor, Ayan Roy · 2026-06-08
arXiv:2606. 06879v1 Announce Type: cross Abstract: Our prior work introduced COVA, a synthetically generated multi-turn conversational smishing dataset of 3,201 labeled conversations, establishing baseline detection benchmarks across eight models.
Read next because An Expanded Synthetic Conversation Dataset for Multi-Turn Smishing Detection overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, under, correct, eval, line, rate. Source: arxiv cs.CR (Cryptography and Security).
arXiv:2606.06879v1 Announce Type: cross Abstract: Our prior work introduced COVA, a synthetically generated multi-turn conversational smishing dataset of 3,201 labeled conversations, establishing baseline detection benchmarks across eight models. While XGBoost with TF-IDF features achieved the best performance, with 72.5\% accuracy and 0.691 macro F1, transformer models underperformed, which was attributed to input truncation and insufficient training data. We present COVA-X, an expanded dataset of 10,985 conversations spanning eight elder-targeted scam categories, produced by an improved generation pipeline addressing contamination, label mismatch, stage-direction bleed, and prompt-design failures from the first iteration. Retraining all classifiers on the expanded dataset yields the central finding of this work: Longformer now surpasses XGBoost on all evaluation metrics, achieving 79.71\% accuracy and 0.7786 macro F1 compared with 78.43\% and 0.7563 for XGBoost. This directly confirms that transformer models require larger conversational corpora to realize their contextual advantages. We additionally document a quality life-cycle including a 12.7$\times$ improvement in label correction rate, from 49.8\% to 3.9\%, an architectural intervention reducing virtual-kidnapping artifact rates from 67.1\% to 46.5\%, and a per-scam-type outcome analysis showing that scam categories modulate results in mechanism-consistent ways. A pre/post-cleanup sensitivity analysis confirms that dataset refinement recovers genuine label-relevant signal across all three classifier architectures.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures, evaluation, benchmark.
- score 100arxiv cs.CL (NLP)arxiv:2606.06865unread
Are Large Language Models Suitable for Graph Computation? Progress and Prospects
Yuting Zhang, Yi Han, Kai Wang, Wei Ni, Angela Bonifati, Wenjie Zhang · 2026-06-08
arXiv:2606. 06865v1 Announce Type: new Abstract: Large language models (LLMs) have been increasingly explored for graph computation, where tasks require reasoning over structured relationships and algorithmic operations.
Read next because Are Large Language Models Suitable for Graph Computation? Progress and Prospects overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, rate, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06865v1 Announce Type: new Abstract: Large language models (LLMs) have been increasingly explored for graph computation, where tasks require reasoning over structured relationships and algorithmic operations. Yet, it remains unclear when LLMs can reliably support such computation and how they should be incorporated into graph-solving pipelines. Existing surveys at the intersection of LLMs and graphs primarily focus on graph learning, text-attributed graphs, or graph-language modeling. To bridge this gap, we provide a comprehensive review of LLMs for graph computation through a role-based taxonomy. Specifically, we identify two major paradigms: i) LLMs as executors, where models directly solve graph tasks from graph descriptions and instructions; and ii) LLMs as planners, where models formulate problems, decompose reasoning steps, and invoke external tools or agents for execution. Based on this taxonomy, we analyze the strengths and limitations of current methods. Our review indicates that LLMs are promising for simple, small-scale tasks, but remain unreliable for large-scale and exactness-demanding tasks. Finally, we summarize available datasets and suggest four future directions.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations.
- score 100arxiv cs.CL (NLP)arxiv:2606.06842unread
CRAFT: A Unified Counterfactual Reasoning Framework for Tabular Question Answering and Fact Verification
Chenshuo Pan, Yu Zhao, Jie Zhang, Changzai Pan, Zhenhe Wu, Jiayi Liang, Yujie Mao, Shuangyong Song, Yongxiang Li, Zhongjiang He · 2026-06-08
arXiv:2606. 06842v1 Announce Type: new Abstract: Table reasoning remains challenging for large language models (LLMs), particularly in tasks that require multi-step inference over long and structured tables.
Read next because CRAFT: A Unified Counterfactual Reasoning Framework for Tabular Question Answering and Fact Verification overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, rate, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06842v1 Announce Type: new Abstract: Table reasoning remains challenging for large language models (LLMs), particularly in tasks that require multi-step inference over long and structured tables. Existing approaches predominantly rely on single-direction reasoning, which limits their ability to explore alternative hypotheses across tasks. In this work, we propose CRAFT, a unified Counterfactual Reasoning Framework that reformulates Tabular question answering and fact verification into a general bidirectional verification process. Our method explicitly constructs both declarative statements and their counterfactual variants. Evidence is then extracted from reasoning along both the original and counterfactual paths, and integrated via a weighted mechanism to arrive at the final answer. Experimental results show that our approach consistently surpasses representative baselines on table reasoning datasets such as WikiTQ and TabFact, achieving especially large improvements on complex question answering. Our framework also significantly mitigates performance gaps between different backbone LLMs. This indicates that counterfactual reasoning effectively overcomes the limitations of single-direction inference, guiding LLMs toward more discerning reasoning and establishing a more principled paradigm for structured reasoning tasks. Our code will be made publicly available upon acceptance.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations.
- score 100arxiv cs.CL (NLP)arxiv:2606.06825unread
Progress-SQL: Improving Reinforcement Learning for Text-to-SQL via Progressive Rewards
Shihao Zhang, Xiaoman Wang, Yuan Liu, Yunshi Lan, Weining Qian · 2026-06-08
arXiv:2606. 06825v1 Announce Type: new Abstract: Reinforcement learning has recently shown promise in improving large language models for Text-to-SQL generation, yet existing methods typically optimize one-shot rewards defined over a single SQL state.
Read next because Progress-SQL: Improving Reinforcement Learning for Text-to-SQL via Progressive Rewards overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, correct, eval, rate, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06825v1 Announce Type: new Abstract: Reinforcement learning has recently shown promise in improving large language models for Text-to-SQL generation, yet existing methods typically optimize one-shot rewards defined over a single SQL state. Such rewards provide limited guidance for iterative SQL correction and are insufficient to capture the improvement of multi-turn SQL refinement. In this paper, we propose Progress-SQL, a multi-turn reinforcement learning framework with progressive rewards for Text-to-SQL. Our approach introduces an Oracle-guided Diagnostic Tree (ODT), which abstracts SQL queries into clause-level structural profiles and produces diagnostic feedback for next-turn refinement. To provide dense and robust reward signals, we combine ODT-based structural alignment with lexical alignment and define a progressive reward that measures the improvement from the initial SQL to the final SQL. We further incorporate a progression latency reward that favors earlier correctness and an execution status reward that encourages recovery from the invalid SQL. Experiments on BIRD, Spider, and Spider robustness variants demonstrate that our method consistently improves Text-to-SQL performance across both primary and robustness evaluations.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses robustness, evaluation.
- score 100arxiv cs.CL (NLP)arxiv:2606.06797unread
Korean Culture into LLM Alignment: Toward Cultural Coherence
MinJae Jung, Minwoo Kim · 2026-06-08
arXiv:2606. 06797v1 Announce Type: new Abstract: Cultural-aspect work on large language models is dominated by a negative target: which outputs to suppress.
Read next because Korean Culture into LLM Alignment: Toward Cultural Coherence overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, line, rate, candidate, capability, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06797v1 Announce Type: new Abstract: Cultural-aspect work on large language models is dominated by a negative target: which outputs to suppress. We argue that a constructive counterpart is also needed, a working definition of what a culturally coherent response is rather than only what it must avoid, and instantiate it for Korean. We design an alignment-data pipeline around a prompt-based LLM seed generator that expands a Korean harm taxonomy, with a Korean-culturally-adapted safe-response policy at its centre: a per-category guideline grounded in Korean legal frameworks, social norms, and interpretive conventions, against which three frontier models each produce a candidate response. DPO fine-tuning on the resulting triplets improves the Korean cultural safe rate across six open-weight LLMs while causing no large degradation on Korean general-capability benchmarks, and qualitative outputs show fine-tuned models naming Korean statutes and institutional procedures and, where appropriate, supplying constructive Korean-context information alongside refusal.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses negative, benchmark.
- score 100arxiv cs.CL (NLP)arxiv:2606.06788unread
Explain Like I'm 5 or Whatever I Choose: Evaluating the Interactive Potential of Language Model Responses
Indu Panigrahi, Tal August · 2026-06-08
arXiv:2606. 06788v1 Announce Type: new Abstract: Evaluations of large language models (LLMs) in scientific information seeking tasks have become increasingly use-centric, such as conducting live or multi-turn evaluations with real users.
Read next because Explain Like I'm 5 or Whatever I Choose: Evaluating the Interactive Potential of Language Model Responses 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, rate, test, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06788v1 Announce Type: new Abstract: Evaluations of large language models (LLMs) in scientific information seeking tasks have become increasingly use-centric, such as conducting live or multi-turn evaluations with real users. These evaluations still assume a single, static chat interface, but as models are integrated into new interfaces, evaluations must shift to incorporate interface-specific criteria. We propose a new evaluation framework based on a formative study with $16$ participants that tests models' ability to generate multiple responses to one query that differ along an interpretable axis of language (language complexity), inspired by direct manipulation interfaces from human-centered design literature. We evaluate GPT-5.1, GPT-5 mini, Claude Sonnet 4.5 + Thinking, and DeepSeek-V3.1 by generating 5 responses at different levels of language complexity for $98$ scientific queries. While models vary complexity across responses, most changes remain inconsistent, with the best performing model (Claude Sonnet 4.5) only shifting reliable complexity measures in the correct direction $46\%$ of the time. Our findings hold with increased sample size and alternative complexity levels.
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.
- score 100arxiv cs.CL (NLP)arxiv:2606.06758unread
A Four-Condition Diagnostic Protocol for Evidence Utilization in Long-Context and Retrieval-Augmented Language Models
Haizhou Xia · 2026-06-08
arXiv:2606. 06758v1 Announce Type: new Abstract: Final-answer accuracy, retrieval recall, and citation overlap do not by themselves identify whether a long-context or retrieval-augmented language model used the evidence it was given.
Read next because A Four-Condition Diagnostic Protocol for Evidence Utilization in Long-Context and Retrieval-Augmented Language Models overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, rate, control, without, full. Source: arxiv cs.CL (NLP).
arXiv:2606.06758v1 Announce Type: new Abstract: Final-answer accuracy, retrieval recall, and citation overlap do not by themselves identify whether a long-context or retrieval-augmented language model used the evidence it was given. A model can answer from parametric memory, fail despite receiving the right passages, or cite evidence without converting it into the requested answer. This paper proposes a matched four-condition evidence-availability protocol--no evidence, full context, retrieved evidence, and oracle-evidence reference--for diagnosing evidence utilization under fixed examples, prompts, score fields, retrieval settings, and validity checks. ONCU is used as a protocol-bound estimator of recovered oracle-reference evidence advantage and is computed only for denominator-valid groups; denominator-free answer, evidence, retrieval, and failure-audit metrics are reported separately. The empirical study evaluates five local open-weight models from the Qwen, Gemma, Llama, and Mistral families across Controlled-ONCU-safe16K, HotpotQA-ONCU, and 2WikiMultiHopQA-ONCU, with 18,000 ONCU-compatible predictions. The main finding is a task-dependent bottleneck split: controlled synthetic settings primarily expose full-context utilization failures, whereas the tested realistic multi-hop settings primarily expose retrieval-chain coverage failures in denominator-free answer and evidence metrics, with ONCU supporting the same direction on oracle-improving groups. The contribution is a diagnostic protocol for separating no-evidence answerability, oracle-evidence recoverability, full-context utilization, and retrieval-conditioned utilization, rather than a single-score leaderboard for long-context or retrieval-augmented systems.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures.
- score 100arxiv cs.CL (NLP)arxiv:2606.06755unread
PromptPrint: Behavioral Biometrics Through Natural Language Prompting in LLMs
Shaiv Patel, Kartik Narayan, Vishal Patel · 2026-06-08
arXiv:2606. 06755v1 Announce Type: new Abstract: Authorship attribution research has traditionally focused on long-form, expressive texts; however, interactions with large language models (LLMs) are typically brief and task-driven prompts.
Read next because PromptPrint: Behavioral Biometrics Through Natural Language Prompting in LLMs overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, under, rate, contexts, language. Source: arxiv cs.CL (NLP).
arXiv:2606.06755v1 Announce Type: new Abstract: Authorship attribution research has traditionally focused on long-form, expressive texts; however, interactions with large language models (LLMs) are typically brief and task-driven prompts. This raises a fundamental question: do such prompts contain a stable, author-identifiable, and distinctive signal? We introduce PromptPrint, a systematic study of prompt-based identity, the hypothesis that a user's habitual vocabulary, syntax, and discourse patterns form a learnable behavioral biometric. Using 20,680 real prompts from 1,034 users, we establish three key findings. First, lexical representations significantly outperform semantic encoders, supporting the "lexical stability hypothesis": identity is primarily encoded in surface-level word choice rather than abstract intent. Second, stylometric features exhibit a "uniqueness-consistency paradox": users are highly distinctive across the population, yet behaviorally inconsistent across contexts. Third, adversarial analysis reveals a clear vulnerability spectrum: identity signals are robust to minor lexical perturbations but degrade substantially under semantic paraphrasing. Overall, our results demonstrate strong identification performance at scale, establishing prompt-based identity as a viable behavioral biometric. This work introduces a new perspective on user modeling in LLM interactions, with important implications for security and privacy. Data and code will be released upon the acceptance of our work.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses adversarial.
- score 100arxiv cs.CL (NLP)arxiv:2606.06738unread
Modular Monolingual Adaptation using Pretrained Language Models
Nalin Kumar, Ond\v{r}ej Du\v{s}ek · 2026-06-08
arXiv:2606. 06738v1 Announce Type: new Abstract: Building monolingual language models (LMs) for low-resource languages typically relies on adapting pretrained language models (PLMs) by finetuning the whole model on the target language.
Read next because Modular Monolingual Adaptation using Pretrained Language Models overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, source, token, rate, full, trained. Source: arxiv cs.CL (NLP).
arXiv:2606.06738v1 Announce Type: new Abstract: Building monolingual language models (LMs) for low-resource languages typically relies on adapting pretrained language models (PLMs) by finetuning the whole model on the target language. This approach is widely favored over training from scratch, as it enables effective knowledge transfer. Additionally, prior work has shown that using a language-specific tokenizer can enhance the adaptability. In this work, we hypothesize that full model tuning is often unnecessary and propose a more modular approach. Specifically, we replace the tokens, freeze the corresponding embeddings, and tune the rest of the model. We use Scottish Gaelic, Irish, and Quechua for our experiments, with Quechua being a very low-resource language (8.5k training instances). Evaluation on natural language understanding (NLU) tasks -- mask filling, NER, and POS -- shows that our proposed approach improves performance when adapting models to low-resource languages. Additionally, we provide a comprehensive analysis of the effectiveness of training strategies, the choice of pretrained embeddings, and models.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation.
- score 100arxiv cs.CL (NLP)arxiv:2606.06715unread
Does Topic Sentiment Cause Perceived Ideology? Comparing Human and LLM Annotations in Political News Articles
Upasana Chatterjee · 2026-06-08
arXiv:2606. 06715v1 Announce Type: new Abstract: We ask whether topic sentiment has a causal effect on perceived political ideology, and whether the answer depends on who assigns the ideology label.
Read next because Does Topic Sentiment Cause Perceived Ideology? Comparing Human and LLM Annotations in Political News Articles 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, line, compare, does, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06715v1 Announce Type: new Abstract: We ask whether topic sentiment has a causal effect on perceived political ideology, and whether the answer depends on who assigns the ideology label. Using articles from AllSides, paired with shared sentiment annotations from Llama-3.3-70b-versatile, we compare ideology labels from expert human annotators, GPT-4o-mini (baseline and finetuned), and Llama-3.3-70B. We apply Double Machine Learning (DML) and community-level mediation analysis across all four annotation paradigms. Human annotations yield no significant causal effects at the community level. Fine-tuned GPT-4o-mini achieves the highest classification accuracy (F1=72.48) and is the only annotator paradigm that produces significant community-level treatment effects and significant natural direct effects (NDEs) in mediation. We interpret this as evidence of shortcut learning: fine-tuning on ideology-labeled data causes the model to internalise a spurious sentiment--ideology coupling not operative in human judgment for this task. This coupling is structurally invisible to F1-based evaluation, with implications for the use of LLM annotations as silver labels and as proxies for human judgment in downstream causal analyses.
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.
- score 100arxiv cs.CL (NLP)arxiv:2606.06708unread
Signal-Driven Observation for Long-Horizon Web Agents
Shubham Gaur, Ian Lane · 2026-06-08
arXiv:2606. 06708v1 Announce Type: new Abstract: Web agents operating over long horizons ingest raw DOM and accessibility trees -- routinely tens of thousands of tokens -- at every action step, causing progressive context degradation that erodes reasoning well before tasks complete.
Read next because Signal-Driven Observation for Long-Horizon Web Agents overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, token, fires, line, full, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06708v1 Announce Type: new Abstract: Web agents operating over long horizons ingest raw DOM and accessibility trees -- routinely tens of thousands of tokens -- at every action step, causing progressive context degradation that erodes reasoning well before tasks complete. We argue that this coupling of observation frequency to action frequency is an architectural mistake. Drawing on the insight from Recursive Language Models that querying a document outperforms reading it wholesale, we propose Signal-Driven Observation (SDO): a dedicated sub-call reads the full DOM but returns only task-relevant elements and their selectors, and is re-invoked only when a lightweight signal detector fires -- triggered by URL transitions, newly visible interactive elements, action failures, or exogenous browser events. We outline the open problems SDO introduces and call on the community to treat observation compression as a core architectural decision in web agent design.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures.
- score 100arxiv cs.CL (NLP)arxiv:2606.06679unread
HKJudge: A Legal Discourse-Annotated Corpus for Interpreting What Courts Find, How They Reason, and What They Rule
Xi Xuan, Wenxin Zhang, Yufei Zhou, King-kui Sin, Chunyu Kit · 2026-06-08
arXiv:2606. 06679v1 Announce Type: new Abstract: Court judgments are central to legal practice and jurisprudence, yet discourse analysis of Hong Kong judgments has received limited attention, owing largely to the absence of expert-annotated corpora.
Read next because HKJudge: A Legal Discourse-Annotated Corpus for Interpreting What Courts Find, How They Reason, and What They Rule overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, source, token, rate, extraction. Source: arxiv cs.CL (NLP).
arXiv:2606.06679v1 Announce Type: new Abstract: Court judgments are central to legal practice and jurisprudence, yet discourse analysis of Hong Kong judgments has received limited attention, owing largely to the absence of expert-annotated corpora. We introduce the Hong Kong Judgment Discourse Dataset (HKJudge), the first sentence-level expert-annotated legal discourse corpus. HKJudge includes criminal judgments across all five levels of HK's court hierarchy, comprising $\sim$290k sentences and $\sim$6.5 million tokens, fully annotated by legal linguistics experts. We design a two-tier discourse schema that captures what facts a court finds, how it reasons, and what it rules. At the sentence level, each sentence is assigned one of 26 rhetorical roles. At the span level, sentences are further annotated with three sentencing elements (charge, imprisonment term, fine). Ten legal linguistics annotators produced the annotations with an inter-annotator agreement of $\kappa = 0.8$. We formulate two tasks on HKJudge, termed rhetorical role classification and legal element extraction, and provide the first benchmark evaluation of four BERT-based models, two open-source LLMs under zero-shot and fine-tuning settings, and four commercial LLMs on both tasks. Our work demonstrates the value of sentence-level discourse annotation for modeling the structure of HK judgments and provides a rich data foundation for future work on legal judgment prediction. The HKJudge dataset and code are available at https://github.com/xuanxixi/HKJudge.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation, benchmark.
- score 100arxiv cs.CL (NLP)arxiv:2606.06674unread
What Do People Actually Want From AI? Mapping Preference Plurality
Julia Sep\'ulveda Coelho, Scott A. Hale · 2026-06-08
arXiv:2606. 06674v1 Announce Type: new Abstract: Large Language Models (LLMs) are often fine-tuned through Reinforcement Learning from Human Feedback (RLHF) to align with people's preferences and values.
Read next because What Do People Actually Want From AI? Mapping Preference Plurality overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, alignment, source, rate, test, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06674v1 Announce Type: new Abstract: Large Language Models (LLMs) are often fine-tuned through Reinforcement Learning from Human Feedback (RLHF) to align with people's preferences and values. However, this method has known limitations: it aggregates conflicting preferences, often relies on unrepresentative samples, and uses only binary comparisons. Analysing 1,500 open-ended responses from the PRISM dataset across 75 countries, we examine what people actually want from AI systems and reveal concrete failures of current methods. We find that different people want different things: most values are requested by fewer than a quarter of respondents, with truthfulness the sole exception at 49%. Furthermore, the same words hide divergent meanings: when people describe what they mean by "truthfulness", they reveal distinct, potentially incompatible, epistemological bases, as some ask for sourced claims, some for expert opinions, and some even ask for unpopular views. Certain capabilities, namely how human-like a model behaves, and some features, like AI guardrails, are outright controversial, with some desiring them and others rejecting them. We additionally find that people often use contextual distinctions (what AI should do "by default" versus "if requested") that binary comparisons cannot capture. These findings expose fundamental problems in current alignment practices. When 49% request truthfulness but define it differently, this is unlikely to be captured by a single reward model. The persistence of high hallucination rates in well-funded models, despite users' clear demands for accuracy, suggests that current methods fail to identify actual preferences. This paper sheds light on the situated, contested, imperfect signals that are currently being flattened into universal preference models, a practice others have characterised as epistemic violence.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses failure, failures, limitation, limitations.
- score 100arxiv cs.CL (NLP)arxiv:2606.06646unread
CAF-Gen: A Multi-Agent System for Enriching Argumentation Structures
Jakub B\k{a}ba, Jaros{\l}aw Chudziak · 2026-06-08
arXiv:2606. 06646v1 Announce Type: new Abstract: Formalizing complex reasoning from natural text is one of the central challenges in computational linguistics.
Read next because CAF-Gen: A Multi-Agent System for Enriching Argumentation Structures overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, word, under, alignment, line, rate, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06646v1 Announce Type: new Abstract: Formalizing complex reasoning from natural text is one of the central challenges in computational linguistics. It requires systems to understand not just keywords but also the context and complex reasoning embedded in a text. Current Argument Mining (AM) techniques identify basic claims and premises, yet they often struggle to capture the richer structural information required by advanced schemas such as the Carneades Argumentation Framework (CAF), which incorporates features such as premise types, proof standards, and argument schemes. We address this limitation by introducing CAF-Gen, an automated multi-agent framework designed to enrich shallow argument structures into CAF-compliant argument models. By employing an iterative Creator-Reviewer pipeline, a creator agent's output is validated by a critical agent to ensure structural integrity. This multi-agent collaboration is crucial for mitigating the structural instability typical of single-pass generative models. Our experiments demonstrate that the iterative feedback loop improves the quality of the resulting data and achieves strong alignment with the original annotations, while producing structurally richer models. Our findings show that the multi-agent system can overcome the limitations of single-pass generation, providing a robust methodology for the automated modeling of formal argumentation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations.
- score 100arxiv cs.CL (NLP)arxiv:2606.06635unread
How Language Models Fail: Token-Level Signatures of Committed and Persistent Reasoning Failures
Tanvi Thoria, Kiana Jafari, Marc R. Schlichting, Mykel J. Kochenderfer · 2026-06-08
arXiv:2606. 06635v1 Announce Type: new Abstract: Failures in language model reasoning emerge through distinct processes that leave identifiable signatures in the reasoning trace.
Read next because How Language Models Fail: Token-Level Signatures of Committed and Persistent Reasoning Failures 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: completions, rect, under, correct, token, rate, full, completion. Source: arxiv cs.CL (NLP).
arXiv:2606.06635v1 Announce Type: new Abstract: Failures in language model reasoning emerge through distinct processes that leave identifiable signatures in the reasoning trace. We characterize these failures using token-level uncertainty signals, finding they arise through two empirically distinguishable processes. The first is committed failure, in which a model locks onto an incorrect reasoning path early in its trace. A central diagnostic signature is the commitment point, beyond which considering additional tokens hurt rather than help failure detection. In the second, persistent uncertainty, uncertainty instead accumulates throughout, and the full trace is needed to best distinguish failing from successful completions. These signatures reproduce across 23 model-dataset configurations, with the framework's falsifiable predictions holding in 20 of 23 cases, well above chance across both failure modes. Finally, we demonstrate our failure mode framework has direct implications for self-consistency, identifying when uncertainty signals complement it and when it can be selectively skipped. These results offer a foundation for understanding when LLM reasoning failures become detectable and for adapting detection strategies accordingly.
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 failure, failures.
- score 100arxiv cs.CL (NLP)arxiv:2606.06622unread
UnpredictaBench: A Benchmark for Evaluating Distributional Randomness in LLMs
Amirhossein Abaskohi, Amirhossein Dabiriaghdam, Liang Luo, Ellie Dingqiao Wen, Lele Wang, Giuseppe Carenini, Peter West · 2026-06-08
arXiv:2606. 06622v1 Announce Type: new Abstract: We introduce UnpredictaBench, an evaluation that tests the ability of large language models (LLMs) to capture true underlying distributions.
Read next because UnpredictaBench: A Benchmark for Evaluating Distributional Randomness in LLMs 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, distributional, eval, rate, capability, test, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06622v1 Announce Type: new Abstract: We introduce UnpredictaBench, an evaluation that tests the ability of large language models (LLMs) to capture true underlying distributions. As LLMs are increasingly used as substitutes for other entities (e.g., for humans in economic simulations), the tendency of many models to collapse towards a single plausible answer means a failure to capture the unpredictability of real systems. Recent work on improving output diversity is insufficient for this setting: simulation requires samples that are calibrated to a target distribution, not merely varied outputs. UnpredictaBench isolates a simplified but fundamental version of this problem: sampling outcomes from individual target distributions, including canonical statistical distributions, distributions induced by stochastic programs, and natural-language scenarios that describe random processes. We introduce 448 such problems together with KS@N, a general-purpose evaluation metric that quantifies how well a model outputs approximate black-box target distributions via the Kolmogorov-Smirnov statistical test. This is the rate at which we fail to reject model samples of size N against ground-truth samples, with larger N indicating greater difficulty. Tested across open and proprietary models, we find a large spread in distributional capabilities. For instance, when models generate samples of size 100 (KS@100, our standard metric), scores range from near 0 to over 20%. No model is able to achieve over 40% at KS@100, showing significant headroom in distributional sampling as a capability. Although adding reasoning can somewhat increase scores, we find no immediate solution for this issue. UnpredictaBench shows that even simple distributional simulation remains challenging, making it a necessary first step toward using LLMs as stand-ins for complex systems.
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, evaluation, benchmark.
- score 100arxiv cs.CL (NLP)arxiv:2606.06614unread
Re-Centering Humans in LLM Personalization
Lechen Zhang, Jiarui Liu, Tal August · 2026-06-08
arXiv:2606. 06614v1 Announce Type: new Abstract: Despite growing interest, most evaluations of large language models' (LLMs') personalization abilities have relied on synthetic data.
Read next because Re-Centering Humans in LLM Personalization overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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 "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: persona, rect, eval, rate, stage, language, model. Source: arxiv cs.CL (NLP).
arXiv:2606.06614v1 Announce Type: new Abstract: Despite growing interest, most evaluations of large language models' (LLMs') personalization abilities have relied on synthetic data. It remains unclear how well current personalization systems work for real users. In this paper, we study the gap in LLM personalization performance when using synthetic versus human data. We collect human conversations (550 conversations) and judgments across three stages of personalization: extracting user attributes from conversations (5,949 judgments), pairing relevant attributes with new prompts (11,919), and incorporating relevant attributes into a personalized response (1,101). Incorporating human data reveals system limitations at each stage. Models struggle to extract attributes from human conversations, disagree with human judgments on relevant attributes, and generate personalized responses that humans judge no better than generic responses (though that LLM judges widely rate as better). We introduce two lightweight training-based interventions that shift automated personalization evaluation closer to human data in our first two stages. However, in our third stage we find that learned reward models achieve only modest correlation with human ratings, suggesting that human-aligned personalization quality judgments are difficult to model directly. Our collected data provides a foundation for studying how models should extract, select, and incorporate user information in ways that humans find useful.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations, evaluation.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06804unread
Interpreting Learning Under Competing Models: Joint and Stepwise Approaches for Dynamic Cognitive Diagnosis
Yawen Ma, Sahoko Ishida, Kate Cain, Gabriel Wallin · 2026-06-08
arXiv:2606. 06804v1 Announce Type: new Abstract: Digital learning environments record learners' responses to individual items, making it possible to study the development of specific skills rather than overall scores.
Read next because Interpreting Learning Under Competing Models: Joint and Stepwise Approaches for Dynamic Cognitive Diagnosis overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06804v1 Announce Type: new Abstract: Digital learning environments record learners' responses to individual items, making it possible to study the development of specific skills rather than overall scores. Drawing conclusions about learning from these data requires a model that links responses to latent skills and tracks how mastery changes over time. When the skills measured by each item are unknown, the analyst must decide whether to estimate this structure, the Q-matrix, jointly with the learning process, or to establish it first and study learning afterwards. We show that this decision can change substantive conclusions about how learners develop. Using dynamic cognitive diagnostic models, we analyse data from two reading games measuring vocabulary and comprehension from Grade 2 to Grade 3, with item-text embeddings providing prior information for the unknown Q-matrix. A joint analysis and a bias-corrected stepwise analysis agree that most learners move toward mastering both skills, but disagree about how many remain only partially proficient at Grade 3, changing how reading progress would be reported. A simulation study identifies when the two analyses diverge and shows that joint analysis is more reliable when the item-skill structure is uncertain and the item pool changes between grades. We provide R code for both analyses.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06776unread
A Rolling-Window Framework for Churn Prediction and Behavioral Driver Identification
Muhammad Jawad Mufti, Omar Hammad, Haitham Saleh, Muqaddas Gull · 2026-06-08
arXiv:2606. 06776v1 Announce Type: new Abstract: Customer churn prediction is a central task in customer analytics, particularly in non-contractual, pay-per-use service environments where disengagement is not explicitly observed and must be inferred from behavioral inactivity.
Read next because A Rolling-Window Framework for Churn Prediction and Behavioral Driver Identification overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, without, alone, full, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06776v1 Announce Type: new Abstract: Customer churn prediction is a central task in customer analytics, particularly in non-contractual, pay-per-use service environments where disengagement is not explicitly observed and must be inferred from behavioral inactivity. Existing churn prediction approaches often rely on simplified temporal assumptions or single-point representations of customer behavior, which limit their ability to support continuous risk assessment, interpretability, and realistic deployment over time. This study proposes a temporally explicit churn prediction framework that models customer behavior using rolling behavioral windows, enabling repeated and instance-level churn risk estimation as customer activity evolves. Customer behavior is summarized within a fixed 30-day observation window, followed by a 30-day future churn evaluation window, ensuring a clear temporal separation between behavioral evidence and churn outcomes. The framework integrates feature-based and sequence-based learning approaches within a unified temporal design. The proposed approach is evaluated on a large-scale, real-world dataset from a non-contractual service platform. Empirical results demonstrate strong and stable predictive performance, with accuracy reaching 87.6% and ROC-AUC of 0.94 for the feature-based model, while the sequence-based model achieves recall as high as 96.1% by capturing temporal disengagement patterns. Evaluation on future unseen data confirms meaningful robustness under temporal shift, with accuracy remaining above 83% and ROC-AUC exceeding 0.91 without model retraining. Overall, the findings highlight that carefully designed temporal framing, rather than model complexity alone, is critical for achieving robust, interpretable, and deployment-ready churn prediction. The study provides a practical foundation for churn-oriented decision support in dynamic service environments.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses robustness, evaluation.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06746unread
Performance Variation in Deep Reinforcement Learning
Haruto Tanaka, A. Rupam Mahmood · 2026-06-08
arXiv:2606. 06746v1 Announce Type: new Abstract: Deep reinforcement learning (RL) algorithms often suffer from low run-to-run robustness, manifesting as significant performance variation across independent runs of identically configured agents.
Read next because Performance Variation in Deep Reinforcement 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 "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, compare. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06746v1 Announce Type: new Abstract: Deep reinforcement learning (RL) algorithms often suffer from low run-to-run robustness, manifesting as significant performance variation across independent runs of identically configured agents. Although this issue poses a spectrum of challenges across research and practice, relatively few studies develop methods to evaluate it; RL research instead often reports uncertainty in the estimated mean performance. In this paper, we outline the limitations of conventional uncertainty and variation estimates, particularly their misalignment with purpose and the risk of underreporting. We then propose an alternative percentile-based statistic and visualization method, min-max IPR and run-wise percentile highlighting, respectively. These percentile-based tools are easy to interpret and rely on standard properties of sample percentiles, providing rich information about run-to-run performance variation. We demonstrate this through three case studies. First, we show that LayerNorm and penultimate-layer normalizations narrow performance variation in PPO, whereas the variation is mostly unchanged in SAC. Second, we compare PPO, SAC, TD-MPC, and TD-MPC2, and show TD-MPC exhibits the least variation while being the most data efficient among the four. Finally, in a comparison of DQN and Rainbow on five Atari environments, we show that both algorithms exhibit similar levels of performance variation.
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, limitations, robustness.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06744unread
Learn to Match: Two-Sided Matching with Temporally Extended Feedback
Haijing Zong, Yancheng Liang, Boyang Zhou, Natasha Jaques · 2026-06-08
arXiv:2606. 06744v1 Announce Type: new Abstract: Two-sided matching markets often involve information that unfolds over time through interviews, repeated interaction, learning, and separation.
Read next because Learn to Match: Two-Sided Matching with Temporally Extended Feedback 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, does, screen, position, lora, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06744v1 Announce Type: new Abstract: Two-sided matching markets often involve information that unfolds over time through interviews, repeated interaction, learning, and separation. Existing matching models typically reduce this process to immediate sub-Gaussian feedback about fixed preferences, missing settings where payoff-relevant information is revealed gradually and changes future matching decisions. We introduce a framework with temporally extended feedback, that formulates two-sided matching as a partially observable Markov game with costly pre-match screening, noisy post-match observations, evolving latent profiles, and endogenous continuation or dissolution. We instantiate this framework in Learn2Match, a multi-agent reinforcement-learning benchmark for dynamic matching markets. Learn2Match supports decentralized decision making over whom to interview, whom to match with, and when to dissolve a match, while evaluating policies using regret, social welfare, and an information-friction loss that measures the welfare gap caused by incomplete revelation of latent preferences. We find that independent PPO achieves higher cumulative social welfare and lower cumulative regret than the bandit-style CA-ETC baseline under temporally extended feedback, demonstrating the promise of MARL for dynamic matching markets. However, PPO still incurs higher information-friction loss, revealing that end-to-end MARL does not yet provide the coordinated exploration structure of matching-bandit methods. These results position Learn2Match as a benchmark for developing the next generation of matching-market algorithms: methods that are adaptive like RL agents, statistically disciplined like bandit algorithms, and structurally aware like stable-matching mechanisms.
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.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06717unread
ShallowBench: Benchmarking Generative Drug Design Models on Shallow-Pocket Targets
Saket Reddy, Shiwei Liu · 2026-06-08
arXiv:2606. 06717v1 Announce Type: new Abstract: While generative AI models have demonstrated remarkable success in structure-based drug design, they predominantly rely on deep binding pockets and struggle to sample effective ligands for challenging low-pocketability targets, such as the historically "undruggable" oncology targets KRAS and MYC.
Read next because ShallowBench: Benchmarking Generative Drug Design Models on Shallow-Pocket Targets 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, alpha, eval, rate, binding, full, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06717v1 Announce Type: new Abstract: While generative AI models have demonstrated remarkable success in structure-based drug design, they predominantly rely on deep binding pockets and struggle to sample effective ligands for challenging low-pocketability targets, such as the historically "undruggable" oncology targets KRAS and MYC. To address this gap, we introduce ShallowBench, a strictly curated benchmark of 5,780 shallow-pocket targets extracted from CrossDocked2020. By computing the difference between an Alpha Shape "lid" volume and the underlying protein atom voxel volume, we successfully isolated targets with low concavity while ensuring sufficient surface area for binding. Evaluating various state-of-the-art generative models reveals weaker predicted binding affinity on these low-concavity interfaces. ShallowBench therefore provides a rigorous benchmark for generative biology models and highlights the necessity of new architectural innovations or loss functions capable of navigating these challenging targets.
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.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06698unread
RECAP: Regression Evaluation for Continual Adaptation of Prompts
Harsh Deshpande, Kushal Chawla, Sangwoo Cho, William Campbell · 2026-06-08
arXiv:2606. 06698v1 Announce Type: new Abstract: Production agentic systems routinely face evolving constraints and must comply from the very next interaction.
Read next because RECAP: Regression Evaluation for Continual Adaptation of Prompts 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, test, model, absent. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06698v1 Announce Type: new Abstract: Production agentic systems routinely face evolving constraints and must comply from the very next interaction. Scenarios like a tool-call notification changing a compliance threshold or a policy update adding disclosure requirements fit this criteria, having close to no room for errors in production. This proactive adaptation setting is common in deployment, but absent from current benchmarks, which assume either static constraint sets or reactive protocols with evaluation feedback. We introduce RECAP, a benchmark that measures continual-learning phenomena (forgetting, regression, forward transfer) at the constraint level under a strictly proactive adapt-then-test protocol: prompt optimization methods receive only the constraint specification and must generalize before seeing any test data. Evaluating six methods across four LLMs and three schedules with evolving constraints, we find that these methods show no significant improvement in performance, even after incurring a higher latency. These methods, designed for offline or reactive settings, are inadequate for the proactive paradigm. Our work emphasizes the growing need for designing proactive prompt adaptation methods, where the models must remain robust to evolving needs in deployment.
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.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06694unread
The Geography of Algorithmic Judgment: LLM Intermediaries, Place Identity, and Racial Steering in Housing Search
Hana Samad, Trung Lam, Christoph M\"ugge-Durum, Michael Akinwumi · 2026-06-08
arXiv:2606. 06694v1 Announce Type: new Abstract: Large language models (LLMs) are rapidly assuming an intermediary role in housing search through the integration of listing platforms within conversational interfaces, mediating access to information, search, and recommendations within urban settings.
Read next because The Geography of Algorithmic Judgment: LLM Intermediaries, Place Identity, and Racial Steering in Housing Search overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, eval, source, line, rate, another. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06694v1 Announce Type: new Abstract: Large language models (LLMs) are rapidly assuming an intermediary role in housing search through the integration of listing platforms within conversational interfaces, mediating access to information, search, and recommendations within urban settings. We expand on prior work on racial steering in LLMs by conducting a behavioral audit of seven open-weight and closed-source LLMs across four U.S. cities, testing location recommendations across three iterative prompting conditions that progressively add lifestyle preference context and reflect fair housing paired-testing methodologies. We find that steering is an emergent behavior of the model's interpretive license rather than primarily a static property. Steering results from the interaction of a user's identity, preference articulation, and the spatial logic that a model has internalized about learned representations of place, preference, and opportunity in a given city, and how different types of users relate to it. While steering was present, it was not uniform in direction or magnitude across evaluated conditions. Preference-conditioned testing often increased or reconfigured the number of models that exhibited steering behaviors relative to baseline conditions, suggesting that LLMs may interpret what the same housing preference means differently depending on the racial identity of the user. Our findings also demonstrate that the city is not a neutral testing unit for LLM evaluation in place-based sectors, and results from one local market cannot be assumed to generalize to another. Local and domain expertise will be required in the housing sector to ensure that legal and institutional commitments to fair housing are not undermined while adopting AI tools that mediate spatial access.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06687unread
Towards Serverless Semi-Decentralized Federated Learning with Heterogeneous Optimizers
Su Wang, Mung Chiang, H. Vincent Poor · 2026-06-08
arXiv:2606. 06687v1 Announce Type: new Abstract: We investigate cluster formation, involving the number and composition of clusters, in decentralized federated learning (FL) with heterogeneous machine learning (ML) optimizers.
Read next because Towards Serverless Semi-Decentralized Federated Learning with Heterogeneous Optimizers 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, full, position, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06687v1 Announce Type: new Abstract: We investigate cluster formation, involving the number and composition of clusters, in decentralized federated learning (FL) with heterogeneous machine learning (ML) optimizers. While clustering in centralized FL has enabled scalability and resource savings, its value and development in fully decentralized environments have yet to be explored. Optimizing cluster formation in such environments is challenging, especially due to the complex coupling between network graph structures, local data heterogeneity, and different local ML model optimizers. To address these challenges, we propose serverless semi-decentralized FL (SSD-FL), a methodology requiring no persistent server infrastructure. In SSD-FL, cluster formation occurs via a lightweight, one-time device-to-device (D2D) initialization phase, after which actual ML model training (alongside consensus and convergence processes) is fully serverless. Functionally, SSD-FL segments global rounds into intra-cluster and inter-cluster regimes, ensuring global convergence and consensus through novel "effective loss functions" that integrate device-specific ML optimizers with network graph-based regularization. Next, SSD-FL leverages the consensus gap via the Cheeger inequality to develop an iterative clustering algorithm evaluated against our derived convergence and consensus bounds, which incorporate a unique scoring metric to quantify data and optimizer heterogeneity across devices. Finally, experimental evaluation against three categories of decentralized FL methodologies validate that SSD-FL improves both convergence speeds and communication efficiency across various network graphs, datasets, and local optimizer regimes.
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 evaluation.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06682unread
Spatiotemporal Imputation with Graph-Informed Flow Matching
Zepeng Zhang, Aref Einizade, Jhony H. Giraldo, Olga Fink · 2026-06-08
arXiv:2606. 06682v1 Announce Type: new Abstract: Missing data is a common challenge in spatiotemporal systems, arising in applications such as air quality monitoring and urban traffic management.
Read next because Spatiotemporal Imputation with Graph-Informed Flow Matching overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, source, rate, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06682v1 Announce Type: new Abstract: Missing data is a common challenge in spatiotemporal systems, arising in applications such as air quality monitoring and urban traffic management. Traditional machine learning approaches, like recurrent and graph neural networks, rely on iterative propagation, which tends to accumulate errors over time and space. Recent diffusion-based methods mitigate error propagation but require iterative sampling and often depend on problem-agnostic Gaussian priors, limiting both efficiency and effectiveness. To address these limitations, we propose GiFlow, a Graph-Informed Flow Matching framework for spatiotemporal imputation. GiFlow replaces the typical Gaussian prior with a graph-informed prior constructed via spatiotemporal filtering of observable signals, which better aligns the source distribution to the target and thereby simplifies the generation trajectory. The flow field is parameterized by a hybrid vector field model that integrates spatial attention, temporal attention, and spatiotemporal propagation, enabling joint modeling of spatial and temporal dependencies. Extensive experiments on both synthetic and real-world datasets demonstrate that the proposed GiFlow outperforms the state-of-the-art approaches in spatiotemporal imputation. The code is available at https://github.com/zepengzhang/GiFlow.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses limitation, limitations.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06673unread
Uncertainty-Aware LLM-Guided Policy Shaping for Sparse-Reward Reinforcement Learning
Ujjwal Bhatta, Utsabi Dangol, Sumaly Bajracharya, Rodrigue Rizk, KC Santosh · 2026-06-08
arXiv:2606. 06673v1 Announce Type: new Abstract: Sparse rewards and heterogeneous task sequences remain persistent challenges in Reinforcement Learning (RL), often resulting in slow convergence, weak generalization, and inefficient exploration.
Read next because Uncertainty-Aware LLM-Guided Policy Shaping for Sparse-Reward Reinforcement Learning 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: eval, line, rate, control, trained, lora, language, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06673v1 Announce Type: new Abstract: Sparse rewards and heterogeneous task sequences remain persistent challenges in Reinforcement Learning (RL), often resulting in slow convergence, weak generalization, and inefficient exploration. We propose Uncertainty-Aware LLM-Guided Policy Shaping (ULPS), a novel framework that integrates a calibrated Large Language Model (LLM) into the RL training loop to provide structured, uncertainty-modulated behavioral guidance. ULPS employs an A*-based oracle to synthesize optimal symbolic trajectories, which are used to fine-tune a BERT-based language model. During training, this model supplies action suggestions whose influence is conditioned on epistemic uncertainty estimated via Monte Carlo (MC) dropout. An entropy-based blending mechanism adaptively balances LLM guidance and the learned policy (via Proximal Policy Optimization, PPO), allowing the agent to prioritize reliable priors while preserving adaptability. We evaluate ULPS on the MiniGridUnlockPickup benchmark and observe consistent improvements in success rate, reward efficiency, and sample complexity over unguided, uncalibrated, and standard RL baselines. ULPS achieves more than 9% improvement in execution accuracy after fine-tuning, requires fewer environment interactions, and yields higher reward AUC. Our results demonstrate that integrating symbolic A* trajectories, pretrained language priors, and uncertainty-aware control offers a principled and effective approach to multi-task reinforcement learning in sparse-reward domains, with potential extensibility to partially observable and multi-agent settings.
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.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06574unread
Skip a Layer or Loop It? Learning Program-of-Layers in LLMs
Ziyue Li, Yang Li, Tianyi Zhou · 2026-06-08
arXiv:2606. 06574v1 Announce Type: new Abstract: Large language models (LLMs) perform inference by following a fixed depth and order, non-recurrent execution of all layers.
Read next because Skip a Layer or Loop It? Learning Program-of-Layers in LLMs 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, eval, rate, trained, language, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06574v1 Announce Type: new Abstract: Large language models (LLMs) perform inference by following a fixed depth and order, non-recurrent execution of all layers. We reveal the wide existence of training-free, flexible, dynamic program-of-layers (PoLar), where pretrained layers can be packed as modules and then skipped or looped to form a customized program for each input. For most inputs, substantially shorter program executions can achieve the same or better accuracy, while incorrect predictions of the original LLM can be corrected by alternative programs with fewer layers. These observations indicate that inference admits multiple valid latent computations beyond the standard forward pass. To efficiently achieve PoLar in practice, we propose a lightweight PoLar prediction network, which learns to generate execution programs that dynamically skip or repeat pretrained layers for each input. Experiments on mathematical reasoning benchmarks demonstrate that PoLar consistently improves accuracy over standard inference and prior dynamic-depth methods, often while executing fewer layers, and that these gains persist under out-of-distribution evaluation. Our results suggest that fixed-depth execution captures only a narrow subset of an LLM's latent reasoning capacity.
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.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06564unread
WAV: Multi-Resolution Block Residual Routing for Deep Decoder-Only Transformers
Kehan Wang · 2026-06-08
arXiv:2606. 06564v1 Announce Type: new Abstract: Residual connections are central to training deep Transformers, but standard PreNorm residual streams aggregate sublayer updates with fixed unit weights.
Read next because WAV: Multi-Resolution Block Residual Routing for Deep Decoder-Only Transformers overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, soft, source, line, language, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06564v1 Announce Type: new Abstract: Residual connections are central to training deep Transformers, but standard PreNorm residual streams aggregate sublayer updates with fixed unit weights. Recent Attention Residuals replace this fixed accumulation with content-dependent depth-wise routing, and Block Attention Residuals make the mechanism efficient by routing over block-level residual summaries. However, a single block summary stores only the low-frequency total residual displacement inside a block, discarding directional structure such as attention-vs-MLP imbalance and early-vs-late block dynamics. We propose WAV v1, a lightweight multi-resolution residual routing method for decoder-only Transformers. Instead of representing each block only by its accumulated residual sum, WAV v1 augments every block with two directional detail bases: a phase basis that contrasts attention and MLP updates, and a split basis that contrasts early and late sublayer updates. These bases are routed together with standard block summaries through the same depth-wise softmax mixer, while negative detail-source initialization and detached RMS matching stabilize training. On character-level TinyStories and Text8 language modeling, WAV v1 shows a clear depth-dependent benefit. Although it is not consistently beneficial at 12 layers, it becomes competitive at 24 layers and outperforms all baselines at 48 layers. At 48 layers, WAV v1 reduces validation loss relative to Block AttnRes from 0.4960 to 0.4738 on TinyStories and from 0.9363 to 0.9305 on Text8, with negligible additional parameters. These results suggest that directional residual details, not only block-level sums, are important for scaling residual routing in deeper Transformers.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses negative.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06560unread
MacArena: Benchmarking Computer Use Agents on an Online macOS Environment
Victor Muryn, Maksym Shamrai, Sofiia Mazepa, Yehor Khodysko · 2026-06-08
arXiv:2606. 06560v1 Announce Type: new Abstract: Computer-use agents (CUAs) operate graphical user interfaces (GUIs) through vision and control primitives, and their capabilities have advanced rapidly, driven in part by standardized online evaluation benchmarks such as OSWorld, which serve both as evaluation tools and as training environments for reinforcement learning.
Read next because MacArena: Benchmarking Computer Use Agents on an Online macOS Environment overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, source, line, rate, control, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06560v1 Announce Type: new Abstract: Computer-use agents (CUAs) operate graphical user interfaces (GUIs) through vision and control primitives, and their capabilities have advanced rapidly, driven in part by standardized online evaluation benchmarks such as OSWorld, which serve both as evaluation tools and as training environments for reinforcement learning. However, macOS remains underserved in this landscape: the only existing benchmark, macOSWorld, covers a narrow slice of first-party applications with simpler tasks, and runs on x86 virtual machines incompatible with Apple Silicon. We introduce MacArena, a benchmark of 421 manually verified tasks spanning 50 applications that combines a curated port of OSWorld tasks, content sourced from macOSWorld, and 49 new macOS-native tasks, all running on Apple's native Virtualization framework on Apple Silicon. We argue that macOS presents distinct GUI challenges beyond what Linux-based benchmarks capture, and our evaluation supports this claim: strong model performance on existing benchmarks can reflect familiarity with task distributions rather than genuine cross-platform GUI competence. Notably, model rankings invert between ported and macOS-native tasks, with a leading model trailing by over 26% on the MacArena subset, suggesting that macOS poses a genuinely harder environment for current GUI agents.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation, benchmark.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06547unread
FAIR-Calib: Frontier-Aware Instability-Reweighted Calibration for Post-Training Quantization of Diffusion Large Language Models
Haoyu Huang, Linlin Yang, Sheng Xu, Boyu Liu, Guodong Guo, Zhongqian Fu, Hang Zhou, Baochang Zhang · 2026-06-08
arXiv:2606. 06547v1 Announce Type: new Abstract: Diffusion Large Language Models (dLLMs) refine tokens iteratively but commit them irreversibly, leading to a "stability lag" where early decisions remain fragile even after being written.
Read next because FAIR-Calib: Frontier-Aware Instability-Reweighted Calibration for Post-Training Quantization of Diffusion Large Language Models overlaps with 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)", experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone". Matching terms: token, line, without, full, stage, position, language, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06547v1 Announce Type: new Abstract: Diffusion Large Language Models (dLLMs) refine tokens iteratively but commit them irreversibly, leading to a "stability lag" where early decisions remain fragile even after being written. We reveal that Post-Training Quantization (PTQ) error easily flips these borderline decisions at the write frontier, which are then permanently locked in and amplified. To address this, we propose Frontier-Aware Instability-Reweighted Calibration (FAIR-Calib), a two-stage PTQ framework for dLLMs. Stage I probes a full-precision teacher to estimate a position prior that combines frontier hits and masked-stage reliability. Stage II performs off-policy, layer-wise calibration by minimizing a reweighted hidden-state MSE, effectively prioritizing the protection of fragile frontier states without requiring expensive end-to-end diffusion rollouts. We further theoretically justify our weighted objective as a surrogate for output KL divergence. Empirically, FAIR-Calib consistently outperforms state-of-the-art baselines on LLaDA and Dream (W4A4), significantly reducing frontier decision flips and suppressing post-commit mismatches across diverse benchmarks.
Potential threat/caveat for 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)": this item discusses benchmark.
- score 100arxiv cs.LG (Machine Learning)arxiv:2606.06546unread
Elmes*: Automated Construction of Fine-Grained Evaluation Rubrics for Large Language Models in Long-Tail Educational Scenarios
Tao Liu, Ye Lu, Ruohua Zhang, Siyu Song, Wentao Liu, Aimin Zhou, Hao Hao · 2026-06-08
arXiv:2606. 06546v1 Announce Type: new Abstract: Evaluating large language models (LLMs) for education requires measuring how models teach, not only what they know.
Read next because Elmes*: Automated Construction of Fine-Grained Evaluation Rubrics for Large Language Models in Long-Tail Educational Scenarios overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, alignment, correct, eval, capability, test, language. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06546v1 Announce Type: new Abstract: Evaluating large language models (LLMs) for education requires measuring how models teach, not only what they know. Existing benchmarks emphasize domain-general correctness or depend on manually designed rubrics that scale poorly to long-tail pedagogical scenarios. We introduce Elmes*, an end-to-end framework for constructing, refining, and applying fine-grained scenario-specific rubrics. Elmes* combines a declarative multi-agent engine for teacher--student--judge interactions with SceneGen, a self-evolving module that co-optimizes evaluation criteria and test data from expert-defined pedagogical dimensions. Using Elmes*, we build Edu-330, covering 330 scenarios across 11 subjects, 3 grade bands, and 10 task types, with over 1{,}000 second-level indicators. Experiments on Edu-330 and four expert-authored gold-standard scenarios show that educational capability is multidimensional: top-tier LLMs differ mainly in creativity and values integration, knowledge-strong models may fail at Socratic scaffolding, and the education-specialized InnoSpark achieves the best human-evaluated average score. LLM judges preserve human-comparable rankings with much lower scoring variance, but exhibit judge-specific biases such as self-preference. Ablations show that expert-scored few-shot anchoring improves human--LLM alignment, while reasoning enforcement and greedy decoding are model-dependent. Elmes* thus provides scalable diagnostic infrastructure for pedagogically grounded LLM evaluation.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses bias, evaluation, benchmark.
- score 100arxiv stat.ML (Machine Learning)arxiv:2505.21423unread
Conflicting Biases at the Edge of Stability: Norm versus Sharpness Regularization
Maria Matveev, Vit Fojtik, Hung-Hsu Chou, Gitta Kutyniok, Johannes Maly · 2026-06-08
arXiv:2505. 21423v3 Announce Type: replace-cross Abstract: The remarkable generalization properties of overparameterized networks are often attributed to implicit biases, such as norm minimization at small learning rates and low sharpness in the Edge-of-Stability regime.
Read next because Conflicting Biases at the Edge of Stability: Norm versus Sharpness Regularization 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, alone, trained, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2505.21423v3 Announce Type: replace-cross Abstract: The remarkable generalization properties of overparameterized networks are often attributed to implicit biases, such as norm minimization at small learning rates and low sharpness in the Edge-of-Stability regime. In this work, we argue that a comprehensive understanding of the generalization performance of gradient descent requires analyzing the interaction between these various forms of implicit regularization. We empirically demonstrate that the learning rate interpolates between low parameter norm and low sharpness of the trained model. We furthermore prove that neither implicit bias alone minimizes the generalization error for diagonal linear networks trained on a simple regression task. These findings demonstrate that focusing on a single implicit bias is insufficient to explain good generalization, and they motivate a broader view of implicit regularization that captures the dynamic trade-off between norm and sharpness induced by non-negligible learning rates.
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.
- score 100arxiv stat.ML (Machine Learning)arxiv:2505.21285unread
Learnable Kernel Density Estimation for Graphs and Its Application to Graph-Level Anomaly Detection
Xudong Wang, Ziheng Sun, Chris Ding, Jicong Fan · 2026-06-08
arXiv:2505. 21285v5 Announce Type: replace-cross Abstract: This work proposes a framework LGKDE that learns kernel density estimation for graphs.
Read next because Learnable Kernel Density Estimation for Graphs and Its Application to Graph-Level Anomaly Detection 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, rate, compare, factor. Source: arxiv stat.ML (Machine Learning).
arXiv:2505.21285v5 Announce Type: replace-cross Abstract: This work proposes a framework LGKDE that learns kernel density estimation for graphs. The key challenge in graph density estimation lies in effectively capturing both structural patterns and semantic variations while maintaining theoretical guarantees. Combining graph kernels and kernel density estimation (KDE) is a standard approach to graph density estimation, but has unsatisfactory performance due to the handcrafted and fixed features of kernels. Our method LGKDE leverages graph neural networks to represent each graph as a discrete distribution and utilizes maximum mean discrepancy to learn the graph metric for multi-scale KDE, where all parameters are learned by maximizing the density of graphs relative to the density of their well-designed perturbed counterparts. The perturbations are conducted on both node features and graph spectra, which helps better characterize the boundary of normal density regions. Theoretically, we establish consistency and convergence guarantees for LGKDE, including bounds on the mean integrated squared error, robustness, and generalization. We validate LGKDE by demonstrating its effectiveness in recovering the underlying density of synthetic graph distributions and applying it to graph anomaly detection across diverse benchmark datasets. Extensive empirical evaluation shows that LGKDE demonstrates superior performance compared to state-of-the-art baselines on most benchmark datasets.
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, benchmark.
- score 100arxiv stat.ML (Machine Learning)arxiv:2509.11208unread
Predictable Compression Failures: Order Sensitivity and Information Budgeting for Evidence-Grounded Binary Adjudication
Leon Chlon, Ahmed Karim, Maggie Chlon, MarcAntonio Awada · 2026-06-08
arXiv:2509. 11208v3 Announce Type: replace Abstract: Transformers used for evidence-grounded binary adjudication (e.
Read next because Predictable Compression Failures: Order Sensitivity and Information Budgeting for Evidence-Grounded Binary Adjudication 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)", experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone". Matching terms: under, token, control, without, length, position, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2509.11208v3 Announce Type: replace Abstract: Transformers used for evidence-grounded binary adjudication (e.g., support/refute, yes/no, or verifier-backed pass/fail decisions) can be sensitive to the order in which exchangeable evidence is presented, producing dispersion across permutations and unreliable attempted answers under a verifier-relative Bernoulli predicate. We treat evidence order as a nuisance variable and formalize an expectation-realization gap: next-token training can minimize expected conditional description length over orderings while a fixed ordering remains position-sensitive. Our Quantified Martingale Violation (QMV) bound predicts the dispersion induced by adjacent-rank positional sensitivity, with $O(\log n)$ growth in the harmonic regime; our Expectation-level Decompression Law (EDFL) specializes a KL convexity/data-processing bound to Bernoulli predicates, yielding Bits-to-Trust (B2T), Risk-of-Hallucination (RoH), and an Information Sufficiency Ratio (ISR) gate for answer/abstain decisions. On 3,059 grounded items from FEVER, HotpotQA, NQ-Open, PopQA, and Controls, we observe logarithmic dispersion and positive Jensen gains from uniform permutation mixtures. In one pre-specified held-out audit (528 items), the analytically fixed ISR$=1$ gate attains 0.0-0.7% hallucination with 20.6-27.9% abstention (95% CIs), supporting the operating point without claiming universal calibration across all model families or unrestricted generation.
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.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.07492unread
Bradley-Terry Rankings for Recommender Systems Across Dataset Taxonomies
Ekaterina Grishina, Stepan Kuznetsov, Askar Tsyganov, Ilya Ivanov, Daria Korovaitceva, Margarita Rusanova, Uliana Parkina, Alexander Derevyagin, Evgeny Frolov, Sergey Samsonov, Anton Lysenko · 2026-06-08
arXiv:2606. 07492v1 Announce Type: cross Abstract: The ranking of recommendation algorithms is a challenging problem since model performance is sensitive to dataset characteristics such as sparsity, sequential structure, and scale.
Read next because Bradley-Terry Rankings for Recommender Systems Across Dataset Taxonomies 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, without, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.07492v1 Announce Type: cross Abstract: The ranking of recommendation algorithms is a challenging problem since model performance is sensitive to dataset characteristics such as sparsity, sequential structure, and scale. This drives a demand for a proper methodology for fair comparison between algorithms. Naive aggregation of performance metrics (e.g., averaging NDCG over benchmarks) can yield misleading rankings, undermining practical selection. To address this problem, we introduce a novel, data-driven ranking methodology based on Bradley-Terry (BT) model. We demonstrate that the obtained ranking depends on key dataset statistics. Additionally, we propose a novel metric for evaluating ranking consistency and demonstrate robustness of our ranking to incomplete data. Finally, we introduce a dataset-specific methodology for ranking algorithms on unseen datasets without running the models, relying on extensions of the Bradley-Terry framework, including BT trees and BT models with covariates.
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, benchmark.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.07483unread
Network Recovery from Cascade Data: A Debiased Jacobian-Based Machine Learning Approach
Lei Huang · 2026-06-08
arXiv:2606. 07483v1 Announce Type: cross Abstract: Many important outcomes unfold as dynamic cascades, including product adoption, disease spread, financial distress, and information diffusion.
Read next because Network Recovery from Cascade Data: A Debiased Jacobian-Based Machine Learning Approach 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, does, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.07483v1 Announce Type: cross Abstract: Many important outcomes unfold as dynamic cascades, including product adoption, disease spread, financial distress, and information diffusion. A central challenge is to recover the hidden influence network behind these cascades. Existing methods typically assume a specific diffusion model, and their performance degrades substantially when that assumption is misspecified. We propose CascadeNet, a Jacobian-based machine learning framework for network recovery that does not require specifying a diffusion mechanism. The key idea is that the underlying influence structure can be characterized by the Jacobian of the one-step transition function. CascadeNet first constructs a flexible estimator of the transition function, and further applies Neyman-orthogonal debiasing via the Riesz representer, so that the debiased Jacobian is $\sqrt{n}$-consistent and asymptotically normal, enabling formal inference on the network structure. We validate CascadeNet in both a simulation exercise and a real-world empirical application. In simulations, where the data-generating process is known, CascadeNet achieves the highest network recovery accuracy across nine common data-generating processes. In an empirical application to COVID-19 transmission across Spain's 52 provinces, CascadeNet recovers transmission networks that are significantly correlated with the true inter-province mobility network, whereas networks recovered by baseline methods show no significant alignment with the ground truth.
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.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.07457unread
Time series Foundation Models based on Physics-Informed Synthetic Histories for Cold-Start Photovoltaic Forecasting
Lorenzo Longarini, Alessandro Rongoni, Simone Silenzi, Emanuele Frontoni, Riccardo Rosati · 2026-06-08
arXiv:2606. 07457v1 Announce Type: cross Abstract: At commissioning time, Photovoltaic (PV) operators must forecast production before target-site observations are available, limiting the direct use of standard supervised forecasters.
Read next because Time series Foundation Models based on Physics-Informed Synthetic Histories for Cold-Start Photovoltaic Forecasting overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, under, eval, source, line, rate. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.07457v1 Announce Type: cross Abstract: At commissioning time, Photovoltaic (PV) operators must forecast production before target-site observations are available, limiting the direct use of standard supervised forecasters. This cold-start setting is addressed with a zero-shot pipeline that generates a synthetic production history from plant metadata and meteorological covariates, enabling time-series foundation models (TSFMs) to forecast through inference-time conditioning. Five TSFMs are benchmarked against classical baselines under strict Cold-Start Baseline, Real Feedback, and Self-Forecast Feedback strategies. The evaluation spans $440$ PV sites across four datasets and diverse climate regimes. Covariate-aware foundation models outperform baselines by approximately $1.7-2\times$: TabPFN-TS achieves the lowest error under Real Feedback (MAE $0.514$, RMSE $0.721$ $kWh$ ${kWp}^{-1}$ ${d}^{-1}$), while Chronos-2 is most robust under Self-Forecast Feedback. Performance is largely insensitive to the synthetic-history source, indicating that accuracy is driven more by the availability of plausible temporal context than by the specific generator.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses evaluation, benchmark.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.07382unread
Covariance Shrinkage via Stochastic Interpolation
Mathieu Chalvidal, Florentin Coeurdoux, Eric Vanden-Eijnden · 2026-06-08
arXiv:2606. 07382v1 Announce Type: cross Abstract: We recast classical shrinkage of high-dimensional covariance estimators as empirical risk minimization over a parametric stochastic interpolant between a source and a target distribution.
Read next because Covariance Shrinkage via Stochastic Interpolation 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, source, line. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.07382v1 Announce Type: cross Abstract: We recast classical shrinkage of high-dimensional covariance estimators as empirical risk minimization over a parametric stochastic interpolant between a source and a target distribution. This formalism recovers known shrinkage estimators as special cases and reveals three distinct mechanisms for reducing statistical risk: (i) Scheduling: the interpolant schedule determines the class of admissible covariances, and hence the achievable risk. (ii) Flow maps and couplings: whereas naive constructions amount to assuming independence between the distributions, specific coupling structures (e.g., solutions of optimal transport problems) can lower the empirical risk. Moreover, non-linear flow maps realizing such couplings free the interpolant covariance from the eigenbasis of the empirical estimate, enabling eigenvector regularization. (iii) Early stopping: estimators defined by integrating a regressed vector field afford an additional bias-variance trade-off through approximation of the true interpolant distribution. We then propose a neural estimator of the interpolant, together with an upper bound on its quadratic risk in terms of the interpolant approximation error, and validate both on synthetic experiments. Finally, we apply the estimator to real neuroimaging data, demonstrating the additional regularization power this approach offers in practice.
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.
- score 100arxiv stat.ML (Machine Learning)arxiv:2606.06742unread
TorchKM: A GPU-Oriented Library for Kernel Learning and Model Selection
Yikai Zhang, Gaoxiang Jia, Jie Ding, Boxiang Wang · 2026-06-08
arXiv:2606. 06742v1 Announce Type: new Abstract: TorchKM is an open-source library for kernel machines, including support vector machines, kernel logistic regression, and kernel quantile regression, with GPU acceleration.
Read next because TorchKM: A GPU-Oriented Library for Kernel Learning and Model Selection overlaps with clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, 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, source, line, full, model. Source: arxiv cs.LG (Machine Learning).
arXiv:2606.06742v1 Announce Type: new Abstract: TorchKM is an open-source library for kernel machines, including support vector machines, kernel logistic regression, and kernel quantile regression, with GPU acceleration. The library features a scikit-learn-style API and is designed to exploit GPU-friendly linear algebra, accelerating the full training and model-selection pipeline through intelligent reuse of matrix operations. Benchmarks show competitive predictive performance together with substantial speedups over standard baselines. Code and documentation are available at https://github.com/YikaiZhang95/torchkm, and the package can be easily installed via PyPI.
Potential threat/caveat for clean result "LoRA persona trained on <A> alone emits <B> at 23.5% when a co-trained partner learns <A>...<B>, vs 0% control on Qwen2.5-7B-Instruct (MODERATE confidence)": this item discusses benchmark.
- score 96arxiv stat.ML (Machine Learning)arxiv:2606.07399unread
Automatic, Debiased, and Invariant Counterfactual Generation under General Interventions
Raphael C Kim, Jingsen Zhu, Ramin Zabih, Michele Santacatterina · 2026-06-08
arXiv:2606. 07399v1 Announce Type: new Abstract: Generative models for counterfactual outcomes have great potential to support decision-making under complex interventions, but existing approaches are limited by unstable estimation, poor generalization across environments, and bias from nuisance model misspecification.
Read next because Automatic, Debiased, and Invariant Counterfactual Generation under General Interventions 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 "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, control, model. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.07399v1 Announce Type: new Abstract: Generative models for counterfactual outcomes have great potential to support decision-making under complex interventions, but existing approaches are limited by unstable estimation, poor generalization across environments, and bias from nuisance model misspecification. We introduce ADIGen, a framework for automatic, debiased, and invariant counterfactual generation under general interventions, including high-dimensional interventions and outcomes. ADIGen combines Riesz regression to avoid unstable density-ratio estimation, causal invariance to improve generalization under distribution shift, and orthogonal statistical learning to obtain doubly robust guarantees against nuisance model misspecification. We provide excess-risk bounds showing that ADIGen controls counterfactual risk under general interventions, with a product-bias nuisance remainder and an invariant risk bound across environments.
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.
- score 96arxiv stat.ML (Machine Learning)arxiv:2606.06785unread
Empirical Transfer Operators and Finite-Sample Change Detection for Noisy Expanding Interval Maps
Aparna Rajput · 2026-06-08
arXiv:2606. 06785v1 Announce Type: new Abstract: We study finite-sample change detection for one-dimensional noisy dynamical systems using partition-based empirical approximations of stationary behaviour.
Read next because Empirical Transfer Operators and Finite-Sample Change Detection for Noisy Expanding Interval Maps 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 "#351 follow-up: broader-vocab position-0 sweep at T=1.0 + position-1 suffix isolation". Matching terms: under, line, rate. Source: arxiv stat.ML (Machine Learning).
arXiv:2606.06785v1 Announce Type: new Abstract: We study finite-sample change detection for one-dimensional noisy dynamical systems using partition-based empirical approximations of stationary behaviour. Given observations from an interval-valued process, we partition the state space, estimate a finite transition matrix from observed transitions between partition elements, and apply a small Doeblin-type regularisation to ensure a unique stationary distribution. From an initial reference segment, we compute a baseline empirical stationary distribution \(\widehat{\pi}_{0,\rho}\). For each later sliding window, we compute \(\widehat{\pi}_{t,\rho}\) and define the score \[ S_t=\|\widehat{\pi}_{t,\rho}-\widehat{\pi}_{0,\rho}\|_1. \] Large values of \(S_t\) indicate a change in stationary behaviour relative to the baseline. The statistic detects changes in invariant density or stationary law, but not all possible changes in transition dynamics. Under explicit assumptions on empirical transition concentration, finite-state stationary distribution stability, partition approximation, regularisation bias, and noise stability, we derive a finite-sample bound for the empirical stationary density. The bound separates sampling error, regularisation bias, partition approximation error, and noise bias. We then obtain a single-window false-alarm guarantee and a sufficient detection condition when the invariant density changes by more than the estimation error. We illustrate the method on synthetic noisy beta-map change-point experiments.
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.
- score 64M7 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.
Read next because Artifact verification caveats for Sagan clean results overlaps with experiment "Add C2 control arm (donor sees marker_B without marker_A) to disambiguate paired-marker binding from marker_B leaking alone". Matching terms: control. Source: M7 QA inline RSS threat source.
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.