AI news archive.
10,659 articles filed under Research, newest first, from the outlets listed on the sources page.
- arXiv cs.AIresearchHow Much Do Activation Monitors Benefit from Original Model Computations?
arXiv:2610.24613v1 Announce Type: cross Abstract: One powerful way to safeguard AI systems is to monitor their internal computations--where signs of undesirable behavior might be more easily detectable than from models'…
- arXiv cs.AIresearchContrastive Adversarial Training for Robust Graph Neural Networks under Label Poisoning
arXiv:2610.24685v1 Announce Type: cross Abstract: Graph Neural Networks (GNNs) are effective for modeling relational data but are vulnerable to label poisoning, where a small number of corrupted training labels can propa…
- arXiv cs.AIresearchTensor Network Datamodels for Training Data Attribution
arXiv:2610.24721v1 Announce Type: cross Abstract: Datamodels predict how a model's behavior changes when different parts of its training data are included or removed. However, linear datamodels assign fixed contributions…
- arXiv cs.AIresearchGLOSS-MT: Controllable Low-Resource Machine Translation with an Editable Dictionary
arXiv:2610.24730v1 Announce Type: cross Abstract: Adapting a pretrained machine translation (MT) model to new terminology usually requires new parallel data and another round of fine-tuning, which is costly for low-resou…
- arXiv cs.AIresearchCorrFlow: Self-Correcting Flow Matching
arXiv:2610.24750v1 Announce Type: cross Abstract: Flow-matching models generate samples through iterative velocity predictions, but typically do not revisit these predictions using the intermediate representations that p…
- arXiv cs.AIresearchEvidence Field: Adaptive Covering Programs for Medical Imaging
arXiv:2610.24773v1 Announce Type: cross Abstract: Medical image classifiers typically return predictions without a native contract for independently reconstructing their spatial decision computation. Evidence Field is a…
- arXiv cs.AIresearchAI Rights with an Off-Switch: Rights-Balancing and Due Process Harmonize AI Rights with AI Safety
arXiv:2610.24782v1 Announce Type: cross Abstract: Several prominent voices in AI safety have recently opposed rights for AI on the grounds that such rights would make it harder for us to justify a mandatory off-switch fo…
- arXiv cs.AIresearchLLM Trojans that Increase Response Lengths and Cost
arXiv:2610.24784v1 Announce Type: cross Abstract: Management of the token-based costs of using LLMs has emerged as a significant problem. Some frontier-model providers charge five times per response token than per prompt…
- arXiv cs.AIresearchWhen LLM-Generated Refactorings Ripple: An Empirical Study of Repository Adoption Readiness
arXiv:2610.24790v1 Announce Type: cross Abstract: Large Language Models (LLMs) are increasingly used to refactor source files, but file-local evaluations do not reveal changes to contracts, dependencies, or caller assump…
- arXiv cs.AIresearchFirst-Order LoRA Training with Phone NPUs
arXiv:2610.24816v2 Announce Type: replace-cross Abstract: On-device fine-tuning lets a large language model (LLM) learn from its user's private data without uploading it. Low-rank adaptation (LoRA) reduces the trainable…
- arXiv cs.AIresearchSpriteStage: Long Horizon Animation Generation via Progressive Scene Programming and Replay
arXiv:2610.24869v1 Announce Type: cross Abstract: Video generation models synthesize visual sequences from text and image inputs. Extending video generation to long animations requires actions, dialogue, camera coverage,…
- arXiv cs.AIresearchRAFT: Redundancy-Aware Input Filtering for Temporal Graph Neural Network Training
arXiv:2610.24875v1 Announce Type: cross Abstract: Temporal graph neural networks (TGNNs) have shown strong performance on dynamic graph learning tasks by continuously updating node memories from evolving interactions (i.…
- arXiv cs.AIresearchLeveraging Mental Models for Human-Centered Explanations: A Systematic Design Approach Illustrated by Explainable Fake Review Identification
arXiv:2610.24903v1 Announce Type: cross Abstract: Explainable AI (XAI) often prioritizes model fidelity over user intelligibility, yielding explanations that satisfy technical transparency yet fail to support human sense…
- arXiv cs.AIresearchHow Well Can LLMs Verbalize Mathematical Formulas? A Multi-Domain Evaluation
arXiv:2610.24917v1 Announce Type: cross Abstract: Large Language Models (LLMs) have shown strong capabilities across natural language tasks. But their ability to generate accurate written verbalizations of mathematical f…
- arXiv cs.AIresearchStable Signer V3: 4D Deformable Primitive Splatting Is All You Need for Sign Language Production
arXiv:2610.24926v2 Announce Type: replace-cross Abstract: We present Stable Signer V3, an interactive high-resolution sign language avatar system built on 4D Deformable Primitive Splatting. It addresses three limitations…
- arXiv cs.AIresearchRobust Direct Preference Optimization under Preference Label-Flip Attacks
arXiv:2610.24945v1 Announce Type: cross Abstract: Direct Preference Optimization (DPO) is widely used for offline Reinforcement Learning from Human Feedback. However, recent works have shown that DPO can be easily manipu…
- arXiv cs.AIresearchControlJEPA: Principled Trajectory Regularization via Lyapunov Tube Loss
arXiv:2610.24951v1 Announce Type: cross Abstract: Next-token prediction leaves hidden-state trajectories largely unconstrained, while existing geometric regularizers primarily control the direction rather than the magnit…
- arXiv cs.AIresearchCasePlay: Self-Play Reinforcement Learning from Case Reports for Medical Reasoning
arXiv:2610.24952v1 Announce Type: cross Abstract: Reinforcement learning (RL) is a promising way to improve the clinical reasoning abilities of large language models (LLMs), but current pipelines often rely on expert-lab…
- arXiv cs.AIresearchLabel-Space Mismatch Distorts Cross-Cohort Evaluation in Breast Cancer Molecular Subtyping
arXiv:2610.24962v1 Announce Type: cross Abstract: Cross-cohort evaluation of breast cancer subtype classifiers is unreliable when source and target cohorts use different annotation standards: METABRIC labels six classes,…
- arXiv cs.AIresearchPRECOCIAL: Pretraining with Code Curriculums via Iterative Search for Accelerated Learning
arXiv:2610.24970v1 Announce Type: cross Abstract: Fully synthetic pretraining data generated by symbolic programs can bootstrap downstream training on naturalistic tasks, such as language modeling, but the generating pro…
- arXiv cs.AIresearchUtoMe: Observation-Uncertainty-Guided Token Merging for Weather Foundation Models
arXiv:2610.25001v1 Announce Type: cross Abstract: Weather transformers process dense reanalysis fields whose reliability varies across variables, locations, and weather regimes. Token merging can reduce computation, but…
- arXiv cs.AIresearchUpLoop: Upcycling Large Language Models with Gated Loop
arXiv:2610.25004v1 Announce Type: cross Abstract: Looped large language models trade parameter storage for recurrent computation, but most are trained from scratch. In practice, pretraining with looped depth is expensive…
- arXiv cs.AIresearchReTrace: Visual Backtracking for Efficient Multimodal Large Language Models
arXiv:2610.25005v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) encode images into dense visual token sequences, creating substantial attention and KV-cache overhead. Existing visual token comp…
- arXiv cs.AIresearchIECLA: Interpretable Emotion Contribution Learning for Affective Image Editing
arXiv:2610.25015v1 Announce Type: cross Abstract: Affective image editing (AIE) aims to evoke a target emotion while preserving the content and structure of a source image. Existing methods improve emotional expression o…
- arXiv cs.AIresearchSynergy-Aware Residual Flow Matching for Single-Cell Perturbation Prediction
arXiv:2610.25021v1 Announce Type: cross Abstract: Predicting single-cell responses to genetic and chemical perturbations is important for understanding regulatory mechanisms and prioritizing therapeutic interventions. Ex…
- arXiv cs.AIresearchOne Head Away from Language Confusion: Single Attention Heads Across Model Families
arXiv:2610.25026v1 Announce Type: cross Abstract: Large language models sometimes reply in an unexpected language, a failure known as language confusion. We ablate individual attention heads in 12 model checkpoints spann…
- arXiv cs.AIresearchHCFM: Helmholtz Component Flow Matching for Time Series Anomaly Detection
arXiv:2610.25059v1 Announce Type: cross Abstract: Flow Matching (FM) is an efficient simulation-free approach for training flow-based generative models, making it a promising foundation for scalable time-series anomaly d…
- arXiv cs.AIresearchPolicy Is Not Capability: Some Safety Benchmarks Measure Which Definition of "Unsafe" You Trained On
arXiv:2610.25060v1 Announce Type: cross 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 e…
- arXiv cs.AIresearchClinSDT: LLM-Encoded Clinical Semantic Guidance for Sepsis Treatment via Decision Transformer
arXiv:2610.25121v1 Announce Type: cross Abstract: Offline sepsis treatment aims to learn treatment policies from retrospective ICU trajectories. Decision Transformer-based methods provide a natural sequence modeling fram…
- arXiv cs.AIresearchTIRE: Top-Down Formula Instantiation with Reliable Execution for Physics Problem Synthesis
arXiv:2610.25134v1 Announce Type: cross Abstract: Advancing physics reasoning in large language models requires scalable training data that connect physical laws, concrete problem scenarios, and verifiable numerical answ…
- arXiv cs.AIresearchReflection-Guided Dynamics Rectification for Text-to-Image Generation
arXiv:2610.25139v1 Announce Type: cross Abstract: Fine-grained rectification of existing generations under a limited additional budget remains a key problem in test-time optimization (TTO) for image generation. As for pr…
- arXiv cs.AIresearchPLANE: Plug-and-Play Fine-Grained Visual Experts for Hallucination Mitigation in Large Vision-Language Models
arXiv:2610.25146v1 Announce Type: cross Abstract: Large Vision-Language Models (LVLMs) commonly suffer from hallucinations, which appear as fabricating non-existent objects or outputting incorrect attributes and spatial…
- arXiv cs.AIresearchLearning without Overwriting: Structural Knowledge Fusion for Continual Learning
arXiv:2610.25150v1 Announce Type: cross Abstract: Catastrophic forgetting in sequential deep learning arises from a fixed-capacity architectural constraint: parameter updates encoding new tasks perturb weight subspaces s…
- arXiv cs.AIresearchM-JEPA: Predictive Self-Supervised Learning for Molecular Graphs with Scaffold-Shift Evaluation on Tox21
arXiv:2610.25163v1 Announce Type: cross Abstract: Self-supervised molecular representation learning is increasingly used to improve transfer on small, label-limited property prediction tasks, yet commonly used contrastiv…
- arXiv cs.AIresearchWorld Models, Trees and Foundation Models for Retail Forecasting: Evidence from Four Public Data Sets
arXiv:2610.25170v1 Announce Type: cross Abstract: A retailer makes different decisions at different grains of the product, store and time hierarchy: store orders for an item in a store, supplier funding for an item acros…
- arXiv cs.AIresearchReranking Answers of Large Language Models with Knowledge Graphs
arXiv:2610.25185v1 Announce Type: cross Abstract: Answering natural language questions over knowledge graph data is challenging due to the vast number of facts, which can be difficult to process and navigate. One potenti…
- arXiv cs.AIresearchViCoR: Estimating Visual Necessity via Counterfactual Residuals for Multimodal Medical Data Selection
arXiv:2610.25202v1 Announce Type: cross Abstract: Medical vision-language models often exhibit hidden stratification, where high overall performance masks failures on subtle visual findings and overlapping clinical condi…
- arXiv cs.AIresearchCalibration-Free Early Stopping for Chain-of-Thought Reasoning via E-processes
arXiv:2610.25212v1 Announce Type: cross Abstract: Reasoning large language models (LLMs) benefit from test-time scaling, motivating adaptive policies that spend tokens when they help and stop early otherwise. Existing ru…
- arXiv cs.AIresearchBatchParScale: Sparse Batch-Mixing for Sequence Models
arXiv:2610.25254v1 Announce Type: cross Abstract: Recent test-time scaling methods improve model performance by increasing inference-time compute. One practical form of this is parallel scaling, where a model samples mul…
- arXiv cs.AIresearchVision as Proprioception: Estimating Humanoid Robot Pose from Video
arXiv:2610.25288v1 Announce Type: cross Abstract: Estimating a humanoid robot's internal state from video has a wide range of real-world applications but remains unexplored. During interactions with a robot, its onboard…
- arXiv cs.AIresearchWorld Model Interpretability: From Decoding Representations to Understanding Latent Worlds
arXiv:2610.25296v1 Announce Type: cross Abstract: Emerging Joint-Embedding Predictive Architectures (JEPAs) are moving beyond representation learning towards latent world models that support prediction, planning, and con…
- arXiv cs.AIresearchLearning When to Abort in Transformer Planning: A Mechanistic Study
arXiv:2610.25300v1 Announce Type: cross Abstract: Planning does not by itself imply the ability to recognise errors, but recognising errors can improve planning capacity. In this paper, we study this phenomenon in Transf…
- arXiv cs.AIresearchShared-Private Cross-Modal Discrepancy Learning for Imbalanced Multi-Source Audio-Visual Deepfake Detection
arXiv:2610.25301v1 Announce Type: cross Abstract: Audio-visual deepfake detectors are often trained under a single dataset or manipulation regime, but multi-source training collections may contain heterogeneous attacks a…
- arXiv cs.AIresearchAdaFedXplain: Adaptive Federated Learning with Explainable AI for Ransomware Detection
arXiv:2610.25318v1 Announce Type: cross Abstract: Attacks using ransomware have developed into complex multiphase campaigns that target vital infrastructure, resulting in annual damages of billions of dollars. Convention…
- arXiv cs.AIresearchWhere and Why Search-Augmented LLMs Fail When Reasoning over Rapidly Evolving Knowledge
arXiv:2610.25323v1 Announce Type: cross Abstract: As the world changes, new information is published while older records remain online. This complicates reasoning for large language models (LLMs), which must retrieve evi…
- arXiv cs.AIresearchFormally Verified Autoresearch for Analyzing the Expressivity of Multi-Head Attention
arXiv:2610.25341v1 Announce Type: cross Abstract: Multi-head attention is a central component of the transformer architecture, yet how its expressivity depends on the number of heads remains poorly understood. We study a…
- arXiv cs.AIresearchLocal Evidence, Global Failure: Diagnosing Speech-Vision Temporal Composition in Audio-Visual Language Models
arXiv:2610.25352v1 Announce Type: cross Abstract: Audio-visual large language models (AV-LLMs) can fail at full-video speech-vision decisions even when useful visual evidence is locally recoverable. We introduce Temporal…
- arXiv cs.AIresearchCHSM8: A Chess Transformer that Follows Natural-Language Style Prompts
arXiv:2610.25356v1 Announce Type: cross Abstract: We present CHSM8, a 210M-parameter chess transformer that plays moves conditioned on a free-text description of a playing style, e.g. "go for the Sicilian Defense", "soli…
- arXiv cs.AIresearchCD-OPD: Contextual Digest for On-Policy Distillation
arXiv:2610.25359v1 Announce Type: cross Abstract: Teacher takeover in on-policy distillation (OPD) can produce incorrect responses even when the student and teacher independently answer the same problem correctly, a fail…
- arXiv cs.AIresearchAligning Cultural Behaviors in Arabic Language Models via Preferences
arXiv:2610.25362v1 Announce Type: cross Abstract: Arabic LLMs have improved rapidly, but aligning them with country-specific cultural behavioral preferences remains underexplored. Previous work has focused primarily on i…