AX·NEWS
Sign in

Artificial Intelligence

Artificial Intelligence news — Wednesday 21 October 2026

150 stories from 11 sources, every one in English. Each links to its publisher.

Open in the live feedRSS feedOn a TV

  1. Broad access through consistent standards
    USOpenAI Blog

    How OpenAI works with distribution partners to uphold consistent safeguards, support responsible access, and clarify expectations in our Usage Policies.

  2. NVIDIA Kumo Relational Outperforms Supervised Machine Learning on Relational Data, No Training Required
    USHugging Face Blog
  3. Train in Simulation, Deploy on a Real Robot: Microduck with Isaac Lab 3.0
    USHugging Face Blog
  4. OpenAI Introduces ‘Cating Lock’ function that hides private conversation on chatGPT
    KRAI Times (Korea) Translated from Korean
  5. COPYRIGHT(C) 2017 SAN DONGJU. ALL RIGHTS RESERVED.
    KRAI Times (Korea) Translated from Korean
  6. Opened the 23rd Gwangju City Farm Festival...Activities and Children's Farming School
    KRAI Times (Korea) Translated from Korean
  7. Participated in the Future Mobility Expo in Gyeongsangnam-do, Republic of Korea
    KRAI Times (Korea) Translated from Korean
  8. Inception, the highest performance dLLM ‘Mercus 3’ released.. . . . . .
    KRAI Times (Korea) Translated from Korean
  9. How Specializing the Vocabulary of SLMs Can Cut Inference Cost by 77% in Real Applications
    USHugging Face Blog
  10. Amazon has expanded its own chip ecosystem. .'Tochi TPU' open source public
    KRAI Times (Korea) Translated from Korean
  11. Face age measurement Beyond Cancer Patient Survival Risk Predict...Hard Researcher, Photo Analysis AI Development
    KRAI Times (Korea) Translated from Korean
  12. Fal Launches ‘Flux 3 Max’ with 22x faster video model than ‘Dance 2.5’
    KRAI Times (Korea) Translated from Korean
  13. How AT＆T is rethinking in-house legal work with OpenAI
    USOpenAI Blog

    AT&T is working with OpenAI to make legal knowledge easier to access. Learn how its LegalEdge team is rethinking intake and scaling legal judgment.

  14. [Bulletin] Flex, Korea MS and Enterprise AI Cooperation
    KRAI Times (Korea) Translated from Korean
  15. Meta ‘Muse’, released 1000 million US downloads in 40 days.. . 12th Record
    KRAI Times (Korea) Translated from Korean
  16. OpenAI acquires “Okestra Bio”, a new drug development management platform
    KRAI Times (Korea) Translated from Korean
  17. China Launches ‘National Data Group’, which integrates AI data resources
    KRAI Times (Korea) Translated from Korean
  18. “Let’s read the health status as a plant electrical signal”...“Plant AI”
    KRAI Times (Korea) Translated from Korean
  19. E-mail:dream@naver.com Copyright(c) HANA FINANCIAL GROUP 2012. ALL RIGHTS RESERVED.
    KRAI Times (Korea) Translated from Korean
  20. Investigation of Self-Surveillance Functions An Unmatched OpenAI Agent...
    KRAI Times (Korea) Translated from Korean
  21. Snaptics, Google ‘Coral’ based edge AI chip open... Wearable and industrial equipment aim
    KRAI Times (Korea) Translated from Korean
  22. Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium.
    USHugging Face Blog
  23. Profiling TorchNeuron (Part 1): System-level profiling on AWS Trainium.
    USHugging Face Blog
  24. Gartner Announces ‘10th Strategy Technology in 2027’
    KRAI Times (Korea) Translated from Korean
  25. Decision Models in Action: Build a Hybrid LLM Workflow
    USTowards Data Science

    A hands-on guide to building browser automation by combining decision models with LLMs. The post Decision Models in Action: Build a Hybrid LLM Workflow appeared first on Towards Data Science .

  26. Meta Reveals Data Center for Learning ‘Muse’...
    KRAI Times (Korea) Translated from Korean
  27. 3.65MWp Solar PV in Spain, Olive field...Remaining and promoting power production
    KRAI Times (Korea) Translated from Korean
  28. Deep throat, AI operational cost reduction core drilling technology 'Deep throat' open source public
    KRAI Times (Korea) Translated from Korean
  29. Granite Speech 5.0: Transcribe, Translate, and Attribute more Accurately and Faster
    USHugging Face Blog
  30. SELECTSTAR, SISCO and AI Reliability Verification
    KRAI Times (Korea) Translated from Korean
  31. "From customer center to sales phone"... Launched ‘Voice Alps’ with voice AI consultation
    KRAI Times (Korea) Translated from Korean
  32. NAVER D2SF invests in OnDevice AI startup tray
    KRAI Times (Korea) Translated from Korean
  33. "Eye" of computer yuzu era AI...Kakao, 2 kinds of vision encoder open source public
    KRAI Times (Korea) Translated from Korean
  34. How to Turn Long-Running Python Simulations Into a Self-Service Platform
    USTowards Data Science

    What I learned about queues, failure recovery, and reproducibility while building asynchronous workloads on AWS The post How to Turn Long-Running Python Simulations Into a Self-Service Platform appeared first on Towards Data Science .

  35. Automated screening under 15kWp in 5 minutes for solar system approval
    KRAI Times (Korea) Translated from Korean
  36. Selected as 7409 of agricultural facilities in Italy...Capital 1 trillion KRW
    KRAI Times (Korea) Translated from Korean
  37. OpenAI, Antropicy IPO Pre-Invested Hurry..."After the Investment Fund"
    KRAI Times (Korea) Translated from Korean
  38. How payroll giant ADP practices AI governance and human oversight
    GBAI Business

    The governance approach focuses on key principles of transparency, explainability, risk management, privacy and security.

  39. ChatGPT Ads expands to 26 new countries as over 100,000 advertisers have used the platform
    USOpenAI Blog

    From local businesses to global brands, more advertisers can now reach people as they explore, compare and decide what to buy in ChatGPT.

  40. AGI Launched Up to the Competitor GPU...Altman “Secure Scenarios”
    KRAI Times (Korea) Translated from Korean
  41. NVIDIA Kumo Forecast Combines Pretrained Time-Series Representations with Historical Pattern Retrieval and Interpretability
    USHugging Face Blog
  42. Spatial Reasoning in AI Models, Taken Apart
    USTowards Data Science

    Grid puzzles that people solve almost perfectly show where language and vision models still break The post Spatial Reasoning in AI Models, Taken Apart appeared first on Towards Data Science .

  43. Developing the 119th structure of the building..."The devotion and sacrifice of the spirit"
    KRAI Times (Korea) Translated from Korean
  44. The Download: introducing the Impossible issue
    USMIT Technology Review Biotech

    This is today’s edition of The Download , our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Introducing: the Impossible issue We wanted our last print edition of 2026 to include a slate of…

  45. MSK, SpaceX, Tesla Merchandise Be the first to comment...Construction of the General Assembly
    KRAI Times (Korea) Translated from Korean
  46. DeepSeek makes DeepInfer open source to cut AI costs
    GBAI Business

    With the move, the Chinese AI vendor is looking to gain an advantage in the open source market, while addressing enterprise concerns about steep token rates.

  47. OpenAI models are coming to Google Cloud
    USOpenAI Blog

    OpenAI and Google Cloud are bringing frontier models and Codex workflows to Gemini’s Model Garden, available through Google Cloud Marketplace.

  48. I Built a Pipeline to Track the Job Market I'm Trying to Break Into.
    USTowards Data Science

    Building an incremental pipeline with dbt snapshots, and the API that quietly lied to me along the way. The post I Built a Pipeline to Track the Job Market I'm Trying to Break Into. appeared first on Towards Data Science .

  49. Announcing SimReady Central: An Open Library of Simulation-Ready 3D Assets for Robotics and AI Factories
    USHugging Face Blog
  50. Overcooking Your Proteins Could Change More Than Its Taste
    SGAsian Scientist

    A new study found that high heat decreased the nutritional content of proteins and was linked to changes in how mice behaved socially.

  51. Cleaning up PFAS
    USMIT Technology Review Biotech

    Scrubbing Earth of per- and polyfluoroalkyl substances—a.k.a. PFAS or forever chemicals—could eliminate a list of woes for people and planet. Used for decades in everything from nonstick pans to firefighting foams, PFAS have been linked to…

  52. We can do impossible things
    USMIT Technology Review Biotech

    I’d wager everyone has a rehearsed response to what’s absolutely one of the world’s worst interview questions: “What’s your biggest weakness?” Whenever I’ve gotten it, I take a slow—even pensive—breath and rattle off my line: “I say no to…

  53. Exercise in a pill
    USMIT Technology Review Biotech

    We all know exercise is good for us. It builds muscle, strengthens bones, lowers blood pressure, and boosts mood. By deciphering how movement benefits the body, researchers hope to develop medicines that can offer some of the same…

  54. Nuclear-powered spaceships
    USMIT Technology Review Biotech

    Just before the Artemis II crew began its historic slingshot around the moon this past April, NASA announced plans to build Space Reactor-1 Freedom. SR-1, the first-ever nuclear-reactor-­powered interplanetary spacecraft, is meant to fly…

  55. Solid-state batteries
    USMIT Technology Review Biotech

    The rise of the lithium-ion battery, which powers everything from the smallest personal electronics to gargantuan energy storage systems, has been nothing short of revolutionary. But as anyone who’s boarded a plane in the last decade could…

  56. [AIReader] Data Analysis Using LLM
    KRAI Times (Korea) Translated from Korean
  57. Interspecies communication
    USMIT Technology Review Biotech

    Humans are far from the only species with a gift for gab, and our furry, feathered, and finned friends could have a lot to tell us. Decoding how animals communicate with one another could offer a window into the inner lives of other…

  58. Eliminating traffic deaths
    USMIT Technology Review Biotech

    Every year, traffic crashes kill more than one million people , and they’re the leading cause of death for children and young adults. In the US, 100 people a day lose their lives this way, in many cases not on highways but on rural and…

  59. Patch the Planet: global response coordination as agents uncover critical vulnerabilities
    USOpenAI Blog

    Patch the Planet connects security researchers, maintainers, and incident responders to help address critical vulnerabilities across open source.

  60. [October 20] AI is the best way to study science...Along with the general work, performance ranking
    KRAI Times (Korea) Translated from Korean
  61. How we built ChatGPT’s novel ad stack for user trust and relevance
    USOpenAI Blog

    Learn how OpenAI built ChatGPT’s LLM-powered ad stack to understand conversational intent, keep ads separate from answers, and deliver relevant ads at scale.

  62. Using thousands of AI agents to uncover new leads in biological data
    USOpenAI Blog

    See how OpenAI, PerturbAI, and Scripps Research used distributed AI agents to uncover promising research leads in data from millions of brain cells.

  63. Rethinking BPE: Merge Selection as Model Selection
    USarXiv cs.CL

    arXiv:2610.26225v1 Announce Type: new Abstract: Subword tokenization methods such as Byte-Pair Encoding (BPE) are a core component of modern language models; yet, tokenizer capacity remains typically determined by a fixed merge budget…

  64. Honeyed Words, Hidden Daggers: Benchmarking and Training LLMs to See Through Mock Politeness
    USarXiv cs.CL

    arXiv:2610.26059v1 Announce Type: new Abstract: The Chinese idiom "kou mi fu jian" ("honeyed words, hidden daggers") names speech that is polite in form but hostile in function. We present ZhenhuanBench, a 1,592-item benchmark built from…

  65. Who Gets Unmasked? Sampler-Induced Bias in Diffusion Language Models
    USarXiv cs.CL

    arXiv:2610.24795v1 Announce Type: new Abstract: Diffusion language models (dLLMs) generate by gradually unmasking a canvas, so their samplers decide not only which token to write but also when to commit it. We show that this second…

  66. Learning from Automatic 3D Gaussian Splatting Edits for Online Bayesian Scene Change Detection
    USarXiv cs.CV

    arXiv:2610.28490v1 Announce Type: new Abstract: Online scene change detection requires integrating incoming observations to determine which regions have changed relative to a reference scene. We update change beliefs as new images arrive…

  67. MLFont: Machine Learning-Driven Font Subsetting and Range-Compact Unicode-Range Slicing for Low-Latency Web Typography
    USarXiv cs.LG

    arXiv:2610.24937v1 Announce Type: new Abstract: CSS3 @font-face rules with unicode-range descriptors partition web fonts into downloadable slices. While Latin fonts need ~20 KB in WOFF2, Pan-European (LGC) and East Asian (CJK) fonts…

  68. Anchoring and Rescaling Attention for Semantically Coherent Inbetweening
    USarXiv cs.CV

    arXiv:2603.17651v2 Announce Type: replace Abstract: Generative inbetweening (GI) seeks to synthesize realistic intermediate frames between the first and last keyframes beyond mere interpolation. As sequences become sparser and motions…

  69. From Code Generator to Engineering Companion: A Plant-aware Agent for PLC Programming
    USarXiv cs.AI

    arXiv:2610.28160v1 Announce Type: cross Abstract: Programmable Logic Controller (PLC) coding remains a manual, IDE-centric task that demands constant context switching between documentation, project files, and code editors. Recent work has…

  70. NLP Meets the World: Toward Improving Conversations With News Media About Natural Language Processing Research
    USarXiv cs.CL

    arXiv:2507.10559v3 Announce Type: replace-cross Abstract: Recent advances in large language models (LLMs) have rapidly increased the public's interest in natural language processing (NLP). This attention is reflected by news outlets…

  71. Matrix-Game 4.0: Harnessing Video Models for Real-Time Playable Game Worlds
    USarXiv cs.CV

    arXiv:2610.27999v1 Announce Type: new Abstract: Moving from controllable videos to genuinely playable worlds requires persistent world state, executable game rules, and low-latency interaction. We outline a roadmap for the next generation…

  72. Deep Generative Crystal Structure Prediction: A Benchmark Study and a Controlled Test of Prototype Dependence
    USarXiv cs.LG

    arXiv:2609.26502v2 Announce Type: replace-cross Abstract: Deep generative models are widely reported to enable de novo crystal structure prediction (CSP), but their capability has not been measured consistently against template-based…

  73. REST: Reliability-Guided Spatial Transcriptomics Supervision for Gene Expression Prediction from Histopathological HE-Stained Images
    USarXiv cs.CV

    arXiv:2610.26988v1 Announce Type: new Abstract: Histopathology is central to cancer diagnosis, but morphology in Hematoxylin and Eosin (HE)-stained images does not directly reveal the molecular states that characterize disease…

  74. Multi-Perspective Evidence Synthesis and Reasoning for Unsupervised Multimodal Entity Linking
    USarXiv cs.CL

    arXiv:2604.20283v2 Announce Type: replace Abstract: Multimodal Entity Linking (MEL) is a fundamental task in data management that maps ambiguous mentions with diverse modalities to the multimodal entities in a knowledge base. However, most…

  75. A Global View of How LLMs Value Human Time
    USarXiv cs.AI

    arXiv:2610.18515v3 Announce Type: replace-cross Abstract: Large language models (LLMs) increasingly act as personal assistants; however, little is known about how they capture national-level preferences on a global scale. Given that "time"…

  76. ValiAssist: Scaling LLM-generated Programming Feedback Validation by Demonstration
    USarXiv cs.AI

    arXiv:2610.27248v1 Announce Type: cross Abstract: Generative AI can create programming feedback at scale, but validating its alignment with instructors' pedagogical goals and strategies remains challenging. Instructors may struggle to…

  77. VEILED-SH: Vulnerability Expression through Indirect Language Evaluation Dataset (Self-Harm)
    USarXiv cs.CL

    arXiv:2610.26190v1 Announce Type: new Abstract: AI systems may miss indirect, culturally situated distress or provide support mismatched to the application context. Existing benchmarks do not jointly address cultural variation, application…

  78. No Escape from the Quadratic Barrier: Adaptive Attacks on Linear $\ell_2$ Sketches
    USarXiv cs.LG

    arXiv:2610.25008v1 Announce Type: cross Abstract: Linear sketches estimate properties of high-dimensional inputs from a fixed low-dimensional projection. At constant accuracy, randomized Euclidean-norm sketches, notably the…

  79. Muon Operates Beyond the Edge of Stability
    USarXiv cs.LG

    arXiv:2610.24305v1 Announce Type: new Abstract: We study the Muon optimizer using sharpness and provide the first empirical evidence that the Flat Minima Hypothesis also applies to Muon. Further, we derive Muon's implicit preconditioner to…

  80. Predictive Closure: A Set-Level Characterization of JEPA's Feature Selection
    USarXiv cs.AI

    arXiv:2610.27364v1 Announce Type: cross Abstract: In joint-embedding predictive architectures (JEPA), an encoder and a predictor are trained together to produce a representation that minimizes the error in predicting the representation of…

  81. The Scheduler Is a Hidden Retriever: How Asynchrony Shapes Evidence Acquisition in Search Agents
    USarXiv cs.CL

    arXiv:2610.26310v1 Announce Type: cross Abstract: Language-model agents increasingly solve complex tasks through parallel reasoning and asynchronous information gathering. Controlled replay across two model families and three benchmarks…

  82. VisualEvolve: A Skill Self-Evolution Framework for Visual Reasoning
    USarXiv cs.AI

    arXiv:2610.28386v1 Announce Type: cross Abstract: Visual tool use enables vision-language models (VLMs) to acquire task-relevant evidence, but its effectiveness can be limited by predefined visual operations and the tool-use strategies…

  83. Beyond Trajectory Statistics: Community-Consistent Human Mobility Generation via Evolving User-Location Graphs
    USarXiv cs.AI

    arXiv:2610.28467v1 Announce Type: new Abstract: Human mobility data are fundamental to Web-based location services and broader urban applications, and mobility generation provides synthetic trajectories when real data are costly or…

  84. MLIPS: Multi-Layer Indexer Pool Sharing
    USarXiv cs.AI

    arXiv:2610.27438v1 Announce Type: cross Abstract: Sparse attention limits the keys used by each query, but finding those keys can still require a full-prefix scan at every layer. We introduce \textbf{MLIPS (Multi-Layer Indexer Pool…

  85. Cross-Lingual Reusability of LLM Pipelines: Financial Extraction, Multilingual Reasoning, and Semantic Screening
    USarXiv cs.AI

    arXiv:2610.27043v1 Announce Type: cross Abstract: Financial institutions increasingly deploy large language models (LLMs) in multilingual markets, raising a practical question: can pipelines developed and validated in English be reused…

  86. LatBridge: Efficient and Stable Model Replacement in Latent-Communication Multi-Agent Systems
    USarXiv cs.AI

    arXiv:2610.27813v1 Announce Type: new Abstract: Multi-agent systems rely on communication to complete tasks, and latent communication passes continuous representations to reduce text generation and re-encoding. However, multi-agent systems…

  87. Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
    USarXiv cs.CL

    arXiv:2607.22583v3 Announce Type: replace-cross Abstract: Large Language Models (LLMs) have achieved widespread adoption because of their strong reasoning and query-response capabilities. However, deploying them in embedded and edge…

  88. Differential privacy representation geometry for medical image analysis
    USarXiv cs.LG

    arXiv:2603.01098v4 Announce Type: replace-cross Abstract: Differential privacy (DP)'s effect in medical imaging is typically evaluated only through end-to-end performance, leaving the mechanism of privacy-induced utility loss unclear. We…

  89. Language steering in latent space to mitigate unintended code-switching
    USarXiv cs.LG

    arXiv:2510.13849v5 Announce Type: replace-cross Abstract: Multilingual Large Language Models (LLMs) often exhibit hallucinations such as unintended code-switching, reducing reliability in downstream tasks. We propose latent-space language…

  90. LandscapeSHAP: Which Persistent Homology Class Gets the Credit?
    USarXiv cs.LG

    arXiv:2609.31469v2 Announce Type: replace-cross Abstract: Shapley values, a solution concept from cooperative game theory, have recently become a standard tool for feature credit allocation in machine learning. They provide an…

  91. Mitigating Data Reconstruction Attacks in Federated Learning: Conditions, Bounds, and Theoretical Guarantees
    USarXiv cs.LG

    arXiv:2407.16735v2 Announce Type: replace-cross Abstract: Privacy attacks, particularly data reconstruction attacks, pose significant challenges to federated learning despite its promise in privacy-preserving collaborative training. This…

  92. Expert L2 English Speech Evaluation on a Gaming PC: Accuracy, Fluency, and Prosody with XLS-R, XEUS, and PhoneticXeus
    USarXiv cs.CL

    arXiv:2610.26406v1 Announce Type: new Abstract: oollim is a locally executable system for evaluating L2 English pronunciation accuracy, fluency, and prosody on a consumer computer. The system combines pretrained XLS-R, XEUS, and…

  93. Answering While You Type: Optimizing Time-to-Answer with Streaming Generation
    USarXiv cs.CL

    arXiv:2610.26296v1 Announce Type: new Abstract: Work on efficient large-language-model inference has concentrated almost entirely on reducing the cost of decoding once a request is complete. The quantity a user experiences, however, is the…

  94. Transport What Shares the Conduction Pathway: Normal-Anchored SPD Geometric Transport for Source-Free Patient Adaptation of Holter and Patch ECG
    USarXiv cs.LG

    arXiv:2610.24430v1 Announce Type: new Abstract: Electrocardiogram (ECG) beat classifiers lose accuracy on new patients and devices, and source-free adaptation methods that assume class-balanced target batches fail on real recordings, whose…

  95. Look Before You Reason: Implicit Visual Thinking for Efficient Multimodal Reasoning
    USarXiv cs.CV

    arXiv:2610.27341v1 Announce Type: new Abstract: Vision-language models (VLMs) have advanced rapidly in multimodal reasoning, yet they remain limited on tasks requiring fine-grained understanding of small or easily overlooked image regions…

  96. Structured Latent Dynamics for Long-Horizon Embodied World Models
    USarXiv cs.LG

    arXiv:2610.24534v1 Announce Type: new Abstract: Reliable long-horizon prediction remains a central challenge for embodied world models. Although recent generative models can synthesize visually plausible action-conditioned futures, their…

  97. LADBench: A Benchmark for Logical Fault Detection in Images
    USarXiv cs.CV

    arXiv:2606.17433v2 Announce Type: replace Abstract: Large Vision Language Models (VLMs) excel at visual question answering and semantic grounding, but their capacity for autonomous logical reasoning remains underexplored. Existing anomaly…

  98. Policy Is Not Capability: Some Safety Benchmarks Measure Which Definition of "Unsafe" You Trained On
    USarXiv cs.LG

    arXiv:2610.25060v1 Announce Type: new Abstract: A guardrail classifier is reported as if unsafe were a property of text. It is not: it is a property of a policy, and the policy enters the model through its labels. We exploit an accident of…

  99. JAXbounce: A Continuous-Control Pong-Like Benchmark for Active Inference Beyond Low Dimensional Environments
    USarXiv cs.AI

    arXiv:2610.28203v1 Announce Type: cross Abstract: We introduce JAXbounce, a continuous control pong-like benchmark designed to evaluate active inference beyond discrete, low dimensional environments. JAXbounce combines continuous states…

  100. ProtGlycanDock: Towards Accurate Protein-Glycan Docking with Tailored Dataset, Benchmark and Model
    USarXiv cs.LG

    arXiv:2610.26009v1 Announce Type: new Abstract: Performing accurate protein-glycan docking is critical to understanding protein-glycan interactions which govern many biological processes. However, this important task is hampered by the…

  101. SLA2: Sparse-Linear Attention with Learnable Routing and Quantization-aware Training
    USarXiv cs.LG

    arXiv:2602.12675v2 Announce Type: replace Abstract: Sparse-Linear Attention (SLA) combines sparse and linear attention to accelerate diffusion models and has shown strong performance in video generation. However, (i) SLA relies on a…

  102. On the Role of Batch Size in Stochastic Conditional Gradient Methods
    USarXiv cs.LG

    arXiv:2603.21191v2 Announce Type: replace Abstract: We study the role of batch size in stochastic conditional gradient methods under a $\mu$-Kurdyka-{\L}ojasiewicz ($\mu$-KL) condition. Focusing on momentum-based stochastic conditional…

  103. Feature Selection for Tabular Foundation Models: Exploring Feature Utility in Low-Data Molecular Prediction
    USarXiv cs.LG

    arXiv:2609.30789v3 Announce Type: replace Abstract: Choosing features is difficult with few labeled molecules, even for a tabular foundation model that learns in context. We study this problem with TabPFN for low-data molecular property…

  104. Recast the Past: Stable Long Video Generation with Full Context Memory
    USarXiv cs.AI

    arXiv:2610.27986v1 Announce Type: cross Abstract: Long-horizon video generation requires retaining an ever-expanding history, but conditioning on its growing token sequence is costly. Existing methods restrict or compress historical…

  105. Factorization-Guided Training and Inference for Diffusion-Drafted Speculative Decoding
    USarXiv cs.LG

    arXiv:2610.23935v1 Announce Type: new Abstract: Speculative decoding accelerates large language model inference through parallel verification of tokens proposed by a drafter. Diffusion drafters further parallelize proposal generation, but…

  106. Provable Adversarial Detection: Prime Quantization meets Gromov--Wasserstein
    USarXiv cs.LG

    arXiv:2610.26102v1 Announce Type: new Abstract: Adversarial vulnerability persists across modern vision architectures from CNNs to vision language models (VLMs), yet existing detection methods rely on heuristics without theoretical…

  107. Two hidden layer suffice: the caterpillar net
    USarXiv cs.LG

    arXiv:2610.26795v1 Announce Type: new Abstract: Every continuous piecewise linear (CPWL) function is a finite signed sum of maxima of affine functions, making maximum representation a central problem in the study of the expressiveness of…

  108. IndusAgent: Agentic Reinforcement Learning for Cross-Category Industrial Anomaly Detection
    USarXiv cs.CV

    arXiv:2605.20682v3 Announce Type: replace Abstract: Small defects in unfamiliar product categories can be difficult to distinguish from benign appearance variations in a global view. Additional reasoning over the same view may leave this…

  109. Adaptive Pullback: Reweighting Supervision at Penalty Jumps in On-Policy Distillation
    USarXiv cs.AI

    arXiv:2610.28194v1 Announce Type: cross Abstract: On-policy distillation (OPD) supervises student-generated responses using sampled-token student-teacher log-ratios. A positive log-ratio penalizes a token that the student assigns higher…

  110. RAFT: Redundancy-Aware Input Filtering for Temporal Graph Neural Network Training
    USarXiv cs.LG

    arXiv:2610.24875v1 Announce Type: new Abstract: Temporal graph neural networks (TGNNs) have shown strong performance on dynamic graph learning tasks by continuously updating node memories from evolving interactions (i.e., events) and…

  111. Fair Regression under Approximate Demographic Parity
    USarXiv cs.LG

    arXiv:2610.25796v1 Announce Type: cross Abstract: Exact demographic parity requires the same prediction distribution across sensitive groups. In regression, this can be too restrictive and reduce predictive accuracy. We instead bound every…

  112. Uncertainty Gates the Gradient: Adaptive Updates for Generalized Few-Shot 3D Segmentation
    USarXiv cs.CV

    arXiv:2610.26909v1 Announce Type: new Abstract: In generalized few-shot 3D point cloud segmentation, support annotations are often compromised by sparsity and occlusion. Blindly adapting to such low quality samples risks learning spurious…

  113. SPLASH: Contrastive Audit Pressure for Detecting and Localizing Security Hallucinations in Generated Code
    USarXiv cs.LG

    arXiv:2610.26001v1 Announce Type: cross Abstract: Code generated by large language models often passes every functional test while remaining insecure. These security hallucinations are common, written with high confidence, and repeated…

  114. RoboJudge: Multimodal Language Models as Judges for Embodied Video Generation
    USarXiv cs.CV

    arXiv:2610.27782v1 Announce Type: new Abstract: Video generation models are promising foundations for embodied world models, but only if their generated rollouts are physically valid and faithful to task instructions. Multimodal large…

  115. Local Conflict, Targeted Repair: A Perturbation Analysis of Multi-Domain RL
    USarXiv cs.CL

    arXiv:2606.02398v2 Announce Type: replace-cross Abstract: Reinforcement learning (RL) post-training improves large language models on individual domains but can degrade previously acquired capabilities. We study this interference in…

  116. GAMF-Net: Grouped Availability-Aware Multimodal Fusion for Smartphone Skin Lesion Classification
    USarXiv cs.CV

    arXiv:2610.28129v1 Announce Type: new Abstract: Smartphone clinical images provide a low-cost data source for computer-aided skin lesion classification in resource-constrained settings, but images alone cannot fully capture patients'…

  117. ABACUS: A Benchmark for Image Editing Models to Assess Object Category and Counting Under Scene Conservation
    USarXiv cs.CV

    arXiv:2610.28316v1 Announce Type: new Abstract: Instruction-based image editing models can now add, remove a specified number of objects or change their category while preserving the scene. Yet these models often fail to produce the…

  118. Auto-Director: Leveraging Multi-Agent Collaboration for Autonomous Video Editing
    USarXiv cs.CV

    arXiv:2610.28346v1 Announce Type: new Abstract: Autonomous video editing from open-world resources remains a challenging problem in multimedia content creation. Retrieval-based methods are constrained by fixed source material and lack…

  119. A Multi-Agent Framework for Automated Machine Learning with Evidence-based Planning for Medical Image Analysis
    USarXiv cs.CV

    arXiv:2610.28410v1 Announce Type: new Abstract: Deep learning has become the default tool for medical image analysis, driven by the open-source movement. However, achieving the best performance on a given task remains difficult, with…

  120. WorldSR: Harnessing World Knowledge Search for Grounded Image Super-Resolution
    USarXiv cs.CV

    arXiv:2610.28423v1 Announce Type: new Abstract: Generative-prior-based super-resolution (SR) methods leverage implicit knowledge learned from large-scale models to enhance low-resolution inputs. However, such priors are inherently…

  121. Not What, But How: A Framework for Auditing LLM Responses across Positioning, Generalization, Anthropomorphism, and Maxims
    USarXiv cs.CL

    arXiv:2606.02493v4 Announce Type: replace Abstract: Large language models (LLMs) are being increasingly used to answer subjective, information-seeking questions, where users are sensitive to how responses are communicated, not just whether…

  122. Touch-JEPA: Tactile Representation Learning from Hand-Object Video with Limited Tactile Supervision
    USarXiv cs.CV

    arXiv:2610.28519v1 Announce Type: new Abstract: Understanding contact and tactile magnitude in hand-object interaction is a key step toward physically intelligent systems, yet learning such representations is limited by the cost and…

  123. S-Agent 2.0: Spatial Reasoning with a General Coding Agent
    USarXiv cs.CV

    arXiv:2610.28354v1 Announce Type: cross Abstract: Spatial reasoning asks a model to work out where objects are in 3D space, how they relate to one another, and how they move. Agentic VLMs have recently brought large improvements on such…

  124. SpanCalib-VLM: Calibrated Hallucination Span Detection in Vision-Language Models
    USarXiv cs.CV

    arXiv:2608.29974v2 Announce Type: replace Abstract: Detecting hallucinations in Large Vision-Language Models (LVLMs) requires both accurate span localization and well-calibrated confidence scores. Fine-tuned generative VLMs excel at…

  125. UltraData-Code: From Raw Repositories to Algorithmically Dense and Task-Oriented Code Data
    USarXiv cs.CL

    arXiv:2610.23960v1 Announce Type: cross Abstract: Code generation has become a core capability of large language models (LLMs), and code data is central to the pre-training process that develops it. As obtaining public code of sufficient…

  126. HumanSkill-4D: Training-Free Language Querying of Animatable Gaussian Avatars in Space and Time
    USarXiv cs.CV

    arXiv:2610.24452v2 Announce Type: replace Abstract: We introduce avatar referring segmentation, which selects regions of animatable Gaussian avatars from language queries. A query can also specify a pose or motion to select matching…

  127. Beyond Prediction: A Component-Level Study of Concept Recovery in In-Context Learning
    USarXiv cs.LG

    arXiv:2610.23936v1 Announce Type: new Abstract: In-context learning (ICL) enables language models to make accurate predictions from demonstrations, but prediction success alone does not reveal the extent to which the underlying concept has…

  128. Can This LLM Annotate My Data? Estimating Zero-Shot LLM Performance
    USarXiv cs.LG

    arXiv:2610.24250v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used in annotation tasks with varying success. While several studies have demonstrated the opportunities and shortcomings of LLMs for…

  129. Whose Credential Is It? How Owner Attribution in Agent MemoryShapes Cross-User Account Misuse
    USarXiv cs.CL

    arXiv:2610.25707v1 Announce Type: cross Abstract: Shared long-term memory lets an LLM agent serving one user inadvertently retrieve another user's credentials, but it is unclear when such exposure leads to actual harm. We introduce a…

  130. A Thousand Faces: A Persuasion Benchmark with Experience-Grounded Persuadees Informed by Psychology and Pragmatics
    USarXiv cs.CL

    arXiv:2610.25206v1 Announce Type: new Abstract: People can express the same beliefs and desires yet need different reasons to change their minds. What makes an argument relevant depends on how those views arose: experiences shape what a…

  131. Superficial Success vs. Internal Breakdown: An Empirical Study of Generalization in Adaptive Multi-Agent Systems
    USarXiv cs.CL

    arXiv:2604.18951v3 Announce Type: replace-cross Abstract: Adaptive multi-agent systems (MAS) are increasingly adopted to tackle complex problems. However, the narrow task coverage of their optimization raises the question of whether they…

  132. MisleadQA: Diagnosing LLM Failures under Misleading Context
    USarXiv cs.CL

    arXiv:2610.24851v1 Announce Type: new Abstract: Large language models (LLMs) are widely used to answer questions from retrieved or user-provided context. However, provided context can be factually true yet out-of-scope, describing a…

  133. Table Serialization Determines Which Relations Language Models Preserve
    USarXiv cs.CL

    arXiv:2610.24798v1 Announce Type: new Abstract: We study how text serialization affects table lookup in instruction-tuned language models. Our diagnostic asks the model to find the row containing a specified value under one named column…

  134. Machina Pauliana: A neuro-symbolic approach to the actio pauliana in Dutch insolvency law
    USarXiv cs.CL

    arXiv:2610.26241v1 Announce Type: cross Abstract: Applying a legal provision frequently requires both structured rule-based reasoning and sensitivity to open-textured terms developed in case law. The actio pauliana of Article 42 of the…

  135. Emergent Sparsity: How Many Effective Variables Survive Coarse-Graining in AlphaEarth Embeddings?
    USarXiv cs.LG

    arXiv:2610.25258v1 Announce Type: new Abstract: AlphaEarth encodes each 10 m land pixel as a 64-component vector intended to support diverse global mapping tasks. Yet many scientific applications work at coarser spatial supports, for which…

  136. Client-Aware Security Adaptive Early Exit Vision Transformers for Heterogeneous Federated Learning
    USarXiv cs.LG

    arXiv:2610.24036v1 Announce Type: new Abstract: Federated learning (FL) trains a shared model across data-owning clients without centralizing raw data, but client data difficulty and compute budgets can differ. We study local early-exit…

  137. PAAC: Privacy-Aware Agentic Device-Cloud Collaboration
    USarXiv cs.CL

    arXiv:2605.08646v2 Announce Type: replace-cross Abstract: Large language model (LLM) agents face a structural tension: cloud agents provide strong reasoning but expose user data, while on-device agents preserve privacy at the cost of…

  138. From Generative to Discriminative: Lightweight Adaptation of SpeechLLMs for Multimodal Emotion Recognition
    USarXiv cs.CL

    arXiv:2609.20081v2 Announce Type: replace Abstract: SpeechLLMs can recognise emotion by generating a label token, but this approach relies on the LLM's existing output representations. We compare this generative approach with a…

  139. DiffuPlan: Editable Planning State for Diffusion Language Model Agents
    USarXiv cs.CL

    arXiv:2610.24717v1 Announce Type: new Abstract: Diffusion language models (dLLMs) are emerging as an alternative to autoregressive (AR) language models, yet dLLM agents largely inherit an interaction protocol designed for AR generation…

  140. Belief Coevolution in a Social Network of Generalist and Specialist Large Language Models: A Controlled Study
    USarXiv cs.CL

    arXiv:2607.27512v2 Announce Type: replace Abstract: Large language models (LLMs) are increasingly deployed in multi-agent environments. However, the processes by which beliefs form and propagate among interacting LLMs remain poorly…

  141. SHINE: A Scalable In-Context Hypernetwork for Mapping Context to LoRA in a Single Pass
    USarXiv cs.CL

    arXiv:2602.06358v4 Announce Type: replace Abstract: We propose SHINE (Scalable Hyper In-context NEtwork), a scalable hypernetwork that can map diverse meaningful contexts into high-quality LoRA adapters for large language models (LLMs). By…

  142. TokenFlow: Latent Token Rectified Flow Matching for 3D Point Cloud Generation
    USarXiv cs.CV

    arXiv:2610.27418v1 Announce Type: new Abstract: 3D point clouds generation requires capturing both part-level layout and surface detail while preserving permutation equivariance. Existing latent generative models often encode an entire…

  143. Text Dominance in Paralinguistic Conflict: Tracing Suppression Through Audio LLM Circuits
    USarXiv cs.CL

    arXiv:2610.25599v1 Announce Type: cross Abstract: Audio LLMs deployed in settings such as crisis hotlines and customer support are only useful beyond speech-to-text pipelines if they use acoustic cues rather than the transcript alone. We…

  144. Dialogue Generation for Conversational Recommendation Using User Search Interactions and Knowledge Graphs
    USarXiv cs.CL

    arXiv:2610.25182v1 Announce Type: cross Abstract: Large-scale conversational recommendation datasets typically require manual dialogue construction or annotation. We propose SearCl, which automatically generates conversational…

  145. LastWords: Boring Problems Revisited
    USarXiv cs.CL

    arXiv:2610.24894v1 Announce Type: new Abstract: In 2022, the second author published a Last Words piece in Computational Linguistics in which he pointed out that, despite the advances in AI up to that point, there were still some problems…

  146. SoK: PIAtlas: Decomposing the Unbounded Prompt Injection Attack Space
    USarXiv cs.CL

    arXiv:2610.24999v1 Announce Type: cross Abstract: Prompt injection is the dominant security risk for large-language-model (LLM) applications and agents, yet the field still lacks a coherent map of how these attacks are constructed. More…

  147. Training Negatives Still Matter After Calibrating Answerability Readouts
    USarXiv cs.CL

    arXiv:2610.26461v1 Announce Type: new Abstract: Answerability readouts predict whether a passage supplies an answer, but whether calibration can compensate for removing negative passages during training remains unclear. We present a…

  148. Region-Guided Food-Package Evidence Extraction and Abstention-Aware Draft R3337 Screening with Selective VLM Review
    USarXiv cs.AI

    arXiv:2610.28211v1 Announce Type: cross Abstract: Applying South Africa's draft R3337 warning criteria to food-package photographs requires nutrition, ingredient and product-form evidence. This study presents a region-guided evidence…

  149. On the Selectivity of Generative Models in Structure-Based Drug Design
    USarXiv cs.LG

    arXiv:2610.24124v1 Announce Type: new Abstract: Designing small-molecule drugs that selectively bind to a target protein while avoiding unintended interactions with other proteins is essential for reducing side effects that contribute…

  150. Reasoning with Context Management and External Memory
    USarXiv cs.CL

    arXiv:2610.24187v1 Announce Type: new Abstract: Reasoning models maintain information about the status of their reasoning process through in-context memory. Complex problems require larger context windows and memory requirements. In this…