AI news archive.
19,854 articles, newest first, from the outlets listed on the sources page.
- arXiv cs.AIagentsDescent-Guided Policy Gradient for Scalable Cooperative Multi-Agent Learning
arXiv:2602.20078v4 Announce Type: replace-cross Abstract: Scaling cooperative multi-agent reinforcement learning (MARL) is fundamentally limited by cross-agent noise. When agents share a common reward, each agent's learn…
- arXiv cs.AIresearchDon't stop me now: How Validation Criteria Affect Checkpoint Selection and Early Stopping
arXiv:2602.22107v2 Announce Type: replace-cross Abstract: Checkpoint selection is a standard component of neural network training, yet the validation criterion used to select a checkpoint is often chosen heuristically. M…
- arXiv cs.AIresearchEfficient Token-Level Policy Optimization for Masked Diffusion Models via Velocity Alignment
arXiv:2603.01563v3 Announce Type: replace-cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) successfully aligns autoregressive models in rigorous domains like mathematical reasoning and code generatio…
- arXiv cs.AIresearchPhysics-Informed Neural Networks with Architectural Physics Embedding for Large-Scale Wave Field Reconstruction
arXiv:2603.02231v2 Announce Type: replace-cross Abstract: Large-scale wave field reconstruction requires precise solutions but faces challenges with computational efficiency and accuracy. The physics-based numerical meth…
- arXiv cs.AIresearchFarther the Shift, Sparser the Representation: Analyzing OOD Mechanisms in LLMs
arXiv:2603.03415v3 Announce Type: replace-cross Abstract: In this work, we investigate how Large Language Models (LLMs) adapt their internal representations when encountering inputs of increasing difficulty, quantified a…
- arXiv cs.AIresearchA theoretical model of dynamical grammatical gender shifting based on set-valued set function
arXiv:2603.03510v3 Announce Type: replace-cross Abstract: This study investigates the diverse characteristics of nouns, focusing on both semantic (e.g., countable/uncountable) and morphosyntactic (e.g., masculine/feminin…
- arXiv cs.AIresearchSalamahBench: Dialect and Category Level Safety Evaluation of Arabic Language Models
arXiv:2603.04410v3 Announce Type: replace-cross Abstract: While different stakeholders are trying to leverage Arabic Language Models (ALMs), safety alignment in ALMs remains largely underexplored, hindering their mainstr…
- arXiv cs.AIagentsThe Trace Is the State: Exact Credit Assignment for LLM Agent Teams
arXiv:2603.06859v4 Announce Type: replace-cross Abstract: Credit assignment for a team of LLM agents, what each message was worth, has mostly been treated as prediction: a learned critic, a trajectory-level score, or an…
- arXiv cs.AIresearchConcept-Guided Fine-Tuning: Steering ViTs away from Spurious Correlations to Improve Robustness
arXiv:2603.08309v3 Announce Type: replace-cross Abstract: Vision Transformers (ViTs) often degrade under distribution shifts because they rely on spurious correlations, such as background cues, rather than semantically m…
- arXiv cs.AIresearchRubiCap: Rubric-Guided Reinforcement Learning for Dense Image Captioning
arXiv:2603.09160v2 Announce Type: replace-cross Abstract: Dense image captioning is critical for vision-language pretraining and for text-to-image generation, but scaling expert-quality annotations is prohibitively expen…
- arXiv cs.AIresearchAdaptive Weighted h-Transform Sampling for Coarse-Guided Visual Generation
arXiv:2603.12057v3 Announce Type: replace-cross Abstract: Coarse-guided visual generation, which synthesizes fine visual samples from degraded or low-fidelity coarse references, is essential for various real-world applic…
- arXiv cs.AIresearchHO-SFL: Hybrid-Order Split Federated Learning with Backprop-Free Clients and Dimension-Free Aggregation
arXiv:2603.14773v3 Announce Type: replace-cross Abstract: Fine-tuning large models on edge devices is severely hindered by the memory-intensive backpropagation (BP) in standard frameworks like federated learning and spli…
- arXiv cs.AIresearchFlashSampling: Fast and Memory-Efficient Exact Sampling
arXiv:2603.15854v3 Announce Type: replace-cross Abstract: Sampling from a categorical distribution is mathematically simple, but in large-vocabulary decoding, it often triggers extra memory traffic and extra kernels afte…
- arXiv cs.AIresearchIndexRAG: Index-Time Reasoning for Multi-Hop Retrieval-Augmented Generation
arXiv:2603.16415v2 Announce Type: replace-cross Abstract: Multi-hop question answering (QA) requires reasoning across multiple documents, yet existing retrieval-augmented generation (RAG) approaches address this either t…
- arXiv cs.AIresearchCompDiff enables fair and zero shot medical image generation across demographic intersections through compositional diffusion
arXiv:2603.16551v4 Announce Type: replace-cross Abstract: Medical image generators trained on imbalanced data can fail at demographic intersections absent from training. We introduce CompDiff, which encodes age, sex and…
- arXiv cs.AIagentsDiscovering What You Can Control: Interventional Boundary Discovery for Reinforcement Learning
arXiv:2603.18257v3 Announce Type: replace-cross Abstract: When an RL agent's observations contain distractors driven by the same confounders as its true state, observational data alone cannot identify which dimensions th…
- arXiv cs.AIresearchWhen Demonstrations Fail: Diagnosing the Limits of In-Context Learning in Large Audio-Language Models with Progressive Cue Removal
arXiv:2603.20433v2 Announce Type: replace-cross Abstract: While Large Audio-Language Models (LALMs) have been shown to exhibit degraded instruction-following capabilities, their ability to infer task patterns from in-con…
- arXiv cs.AIresearchBenchmarking Bengali Dialectal Bias: A Multi-Stage Framework Integrating RAG-Based Translation and Human-Augmented RLAIF
arXiv:2603.21359v2 Announce Type: replace-cross Abstract: Large language models (LLMs) frequently exhibit performance biases against regional dialects of low-resource languages. However, frameworks to quantify these disp…
- arXiv cs.AIresearchP^2O: Joint Policy and Prompt Optimization
arXiv:2603.21877v4 Announce Type: replace-cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) enhances Large Language Model (LLM) reasoning but is suffer from advantage collapse: when all rollouts of a…
- arXiv cs.AIagentsSelective Deficits in LLM Mental Self-Modeling in a Behavior-Based Test of Theory of Mind
arXiv:2603.26089v3 Announce Type: replace-cross Abstract: The ability to represent oneself and others as agents with knowledge, intentions, and belief states that guide their behavior - Theory of Mind - is a human univer…
- arXiv cs.AIresearchThe Model Says Walk: Measuring whether LLMs Condition on Hidden Constraints
arXiv:2603.29025v4 Announce Type: replace-cross Abstract: Asked whether to walk or drive to a car wash 50 m away, most language models say walk, forgetting that the car has to be there. Such failures are usually measured…
- arXiv cs.AIagentsMeta-TTL: Meta-Learning Self-Improvement Policies for Language Agents
arXiv:2604.00830v4 Announce Type: replace-cross Abstract: Test-Time Learning (TTL) enables language agents to iteratively refine their performance through repeated interactions with the environment at inference time. At…
- arXiv cs.AIresearchAromaGen: Interactive Generation of Rich Olfactory Experiences with Multimodal Language Models
arXiv:2604.01650v4 Announce Type: replace-cross Abstract: Smell's connections with food, memory, and social experience have motivated researchers to bring olfaction into interactive systems. Multimodal AI opens new possi…
- arXiv cs.AIresearchAn Inspectable LLM Council for Multi-Model Research Answer Aggregation
arXiv:2604.02923v5 Announce Type: replace-cross Abstract: Long-form research questions often require a single answer that combines details, reconciles conflicting estimates, and retains qualifications spread across sever…
- arXiv cs.AIresearchHow Does "English (US)" Become the Default? Triangulating Structural Bias Towards American English Across the LLM Pipeline
arXiv:2604.04204v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are increasingly embedded in educational, professional, and public infrastructure, yet widely used platforms expose "English (US)" as…
- arXiv cs.AIresearchRelative Density Ratio Optimization for Stable and Statistically Consistent Model Alignment
arXiv:2604.04410v2 Announce Type: replace-cross Abstract: Aligning language models with human preferences is essential for ensuring their safety and reliability. Although most existing approaches assume specific human pr…
- arXiv cs.AIresearchEffiPair: Improving the Efficiency of LLM-generated Code with Differential Execution Feedback
arXiv:2604.05137v2 Announce Type: replace-cross Abstract: Large language models (LLMs) can generate functionally correct programs that differ substantially in execution efficiency. Existing inference-time optimization me…
- arXiv cs.AIresearchAn Imperfect Verifier is Good Enough: Learning with Noisy Rewards
arXiv:2604.07666v2 Announce Type: replace-cross Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) is widely used for post-training Large Language Models, but practical verifiers can make errors. We study ho…
- arXiv cs.AIresearchHiFloat4 Format for Language Model Pre-training on Ascend NPUs
arXiv:2604.08826v2 Announce Type: replace-cross Abstract: Training large foundation models at low numerical precision is one of the most promising directions for reducing the compute and memory cost of modern AI. Recent…
- arXiv cs.AIresearchFastGrasp: Learning-based Whole-Body Control Method for Fast Dexterous Grasping with Mobile Manipulators
arXiv:2604.12879v2 Announce Type: replace-cross Abstract: Fast dexterous grasping while a mobile base remains in motion requires coordinated whole-body control and rapid adaptation to physical contact. We propose FastGra…
- arXiv cs.AIresearchLightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation
arXiv:2604.13010v3 Announce Type: replace-cross Abstract: On-policy distillation (OPD) is an effective post-training paradigm for large language models but requires a live teacher server throughout training, resulting in…
- arXiv cs.AIresearchUniRect-CoT: Enhancing Generation in Unified Multimodal Models via Reflective Rectification with Inherent Understanding
arXiv:2604.13540v2 Announce Type: replace-cross Abstract: Unified Multimodal Models (UMMs) aim to integrate visual understanding and generation within a single structure. However, these models exhibit a notable capabilit…
- arXiv cs.AIresearchMambaSL: Exploring Single-Layer Mamba for Time Series Classification
arXiv:2604.15174v3 Announce Type: replace-cross Abstract: Despite recent advances in state space models (SSMs) such as Mamba across various sequence domains, research on their standalone capacity for time series classifi…
- arXiv cs.AIresearchWhy Fine-Tuning Encourages Hallucinations and How to Fix It
arXiv:2604.15574v3 Announce Type: replace-cross Abstract: Large language models are prone to hallucinating factually incorrect statements. A key source of these errors is exposure to new factual information through super…
- arXiv cs.AIresearchDemystifying the Unreasonable Effectiveness of Greedy Alignment Methods
arXiv:2604.17207v2 Announce Type: replace-cross Abstract: Greedy AI alignment methods based on Bradley-Terry reward-model fitting and KL-regularized Alignment are remarkably effective in both offline and online preferenc…
- arXiv cs.AIresearchLearning Evidence Highlighting for Frozen LLMs
arXiv:2604.22565v3 Announce Type: replace-cross Abstract: Large Language Models (LLMs) can reason well, yet often miss decisive evidence when it is buried in long, noisy contexts. We introduce HiLight, an Evidence Emphas…
- arXiv cs.AIresearchSGP-SAM: Self-Gated Prompting for Transferring 3D Segment Anything Models to Lesion Segmentation
arXiv:2604.22825v2 Announce Type: replace-cross Abstract: Large segmentation foundation models such as the Segment Anything Model (SAM) have reshaped promptable segmentation in natural images, and recent efforts have ext…
- arXiv cs.AIresearchDRAGON: A Benchmark for Evidence-Grounded Visual Reasoning over Diagrams
arXiv:2604.25231v2 Announce Type: replace-cross Abstract: Diagram question answering (DQA) requires models to interpret structured visual representations such as charts, maps, infographics, circuit schematics, and scient…
- arXiv cs.AIresearchVUDA: Enabling Controlled Spatial Sharing of Graphics and Compute on NVIDIA GPUs
arXiv:2605.01352v2 Announce Type: replace-cross Abstract: Graphics and compute increasingly share a single GPU in embodied AI simulators, AI-enabled games, and VR systems. Running these workloads concurrently can improve…
- arXiv cs.AIresearchQuery-Dependent Use of Generated Descriptions for Reliable Visual Question Answering
arXiv:2605.01733v5 Announce Type: replace-cross Abstract: Vision-Language Models (VLMs) hallucinate objects that are not present, and a growing line of work tries to curb this by feeding the model its own generated capti…
- arXiv cs.AIresearchFederated Semi-Supervised Graph Neural Networks with Prototype-Guided Pseudo-Labeling for Privacy-Preserving Gestational Diabetes Mellitus Prediction
arXiv:2605.01810v2 Announce Type: replace-cross Abstract: Gestational Diabetes Mellitus (GDM) is a high-prevalence pregnancy complication that requires accurate early risk stratification to reduce maternal and fetal morb…
- arXiv cs.AIresearchRoboAlign-R1: Distilled Multimodal Reward Alignment for Robot Video World Models
arXiv:2605.03821v2 Announce Type: replace-cross Abstract: Existing robot video world models are typically trained with low-level objectives such as reconstruction and perceptual similarity, which are poorly aligned with…
- arXiv cs.AIagentsDeco: Extending Cherished Physical Objects into AI Companion Agents through Dual Embodiment
arXiv:2605.03882v2 Announce Type: replace-cross Abstract: Physical objects (e.g., plush toys) can transcend materiality to become emotional anchors and provide companionship. However, these bonds remain one-sided because…
- arXiv cs.AIagentsMOSAIC-Bench: Measuring Compositional Vulnerability Induction in Coding Agents
arXiv:2605.03952v2 Announce Type: replace-cross Abstract: Coding agents often pass per-prompt safety review yet ship exploitable code when their tasks are decomposed into routine engineering tickets. The challenge is str…
- arXiv cs.AIresearchDemystifying Manifold Constraints in LLM Pre-training
arXiv:2605.04418v2 Announce Type: replace-cross Abstract: The recent success of matrix optimizers (e.g., Muon) suggests that specific normalization of momentum, such as orthogonalization and row-wise normalization, benef…
- arXiv cs.AIagentsEvolving Idea Graphs with Learnable Edits-and-Commits for Multi-Agent Scientific Ideation
arXiv:2605.04922v2 Announce Type: replace-cross Abstract: LLM-empowered multi-agent systems offer new potential to accelerate scientific discovery by generating novel research ideas. However, existing methods typically c…
- arXiv cs.AIresearchTwo-Stage Learned Decomposition for Scalable Routing on Multigraphs
arXiv:2605.05389v2 Announce Type: replace-cross Abstract: Most neural methods for Vehicle Routing Problems (VRPs) are limited to Euclidean settings or simple graphs. In this work, we instead consider multigraphs, where p…
- arXiv cs.AIresearchOne Turn Too Late: Learning When to Intervene Against Multi-Turn Malicious Intent
arXiv:2605.05630v3 Announce Type: replace-cross Abstract: Hidden malicious intent in multi-turn dialogue poses a growing threat to deployed large language models (LLMs). Rather than exposing a harmful objective in a sing…
- arXiv cs.AIresearchSymDrift: One-Shot Generative Modeling under Symmetries
arXiv:2605.06140v3 Announce Type: replace-cross Abstract: Generative modeling of physical systems, such as molecules, requires learning distributions that are invariant under global symmetries, such as rotations in three…
- arXiv cs.AIagentsPushing the accuracy of on-top functionals with agent-driven supervised learning
arXiv:2605.06215v3 Announce Type: replace-cross Abstract: Multiconfiguration pair-density functional theory (MC-PDFT) combines multiconfigurational reference wave functions with on-top functionals to provide an efficient…