Papers

New on arXiv

Sat 3 Oct 2026

323 papers

BoT-Feedback: Grounding Multimodal Reasoning in Biomechanical Evidence for Explainable Human Action FeedbackLLM reasoning → Multimodal reasoning +1
AegisFlow: A Multi-Agent Agentic AI Framework for Autonomous Remediation and Self-Healing in Fragile Data EcosystemsLLM agents and tool use → Software engineering coding agents
APEX: Active Protection at Execution Boundaries for LLM AgentsLLM agents and tool use → Agent threats, guardrails and governance +1
ACG-WAM: World-Action Modeling via Action-Conditioned Geometric Latent PredictionWorld models → Embodied and robot world models
Principles that Guide, Actions that Inform: Agent Evolution via Knowledge AbstractionLLM agents and tool use → General LLM agent frameworks and benchmarks
Learning to Decide, Not to Reason: Parameter-Efficient Decision Operators via Low-Rank Activation SteeringLLM reasoning → Reinforcement learning for reasoning +1
AdaLoop: Adaptive-Depth Latent Reasoning for Audio Language ModelsLLM reasoning → Audio-language model reasoning
Beyond the Linear Representation Hypothesis: Non-Linear Activation Steering in Text-to-Image ModelsMechanistic interpretability → Interpreting and steering diffusion models
Learning to Remember: Distilling Memory Retention for Compact Recurrent Neural NetworksKnowledge distillation → Distillation for continual learning
Stabilizing language models under continual learning via condition-anchored distillationDiffusion models → Discrete and text diffusion
RADC: Risk-Aware Dual Caching for Vision-Language Test-Time AdaptationMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
Near-Optimal Sample Complexity for Recursive Entropic Risk Reinforcement Learning with a Generative ModelDeep reinforcement learning methods → RL theory and sample complexity
AttSVD:Prompt-Adaptive Low-Rank KV Cache Compression via Attention-Guided SVDEfficient LLM inference and serving → KV cache optimization
SWAP: Stepwise Action Policy Routing for Vision-Language-Action ModelsVision-language-action models → Reinforcement learning and post-training
DriftSR: One-Step Real-World Image Super-Resolution via Distribution DriftingDiffusion models → Inverse problems and image restoration
ExStereo: Lifting 2D Vision-Language-Action Models to 3D with Explicit Stereo RepresentationsVision-language-action models → 3D and spatial-aware VLA
MedImageOSWorld: Benchmarking GUI Agents for Medical Image ConsolesLLM agents and tool use → Medical and scientific domain agents
Knowing the Store: What a Memory Backend Must Write Down Before an Agent Can Read ItLLM agents and tool use → Agent memory systems
Pressure, Context, and Machine Self-Control: A Criminological Test of Reward Hacking in Generative AI ModelsLLM alignment / RLHF / preference optimisation → Reward hacking and misspecification
Repeated-Measure Leakage, Distribution Shift, and Reliability under Partial Observation in Patient World ModelsWorld models → Medical and surgical world models
ARISE: Adaptive Agentic Reasoning with Image-grounded Self-Evaluation for Interpretable IBD AssessmentLLM reasoning → Clinical and biomedical applications +1
Learning Task and Motion Plans from Real Demonstrations with Hybrid Flow MatchingDiffusion models → Diffusion and flow policies for robotics
Flow Policies as Actions of Skill-Level World Models: Learned and Symbolic Abstractions for Long-Horizon PlanningWorld models → Embodied and robot world models
PatternDex: Learning Interaction Patterns to Guide Reinforcement Learning of Bimanual Dexterous Manipulation of Articulated ObjectsDeep reinforcement learning methods → RL for robot locomotion and manipulation
When Is Enough Enough in Self-Evolving LLM Systems?Synthetic data and self-improvement for LLMs → Self-rewarding alignment and self-training
FoSeRL: Formal Sequential Robustness Certification for Reinforcement Learning PoliciesDeep reinforcement learning methods → Safe and constrained RL
Variance-Aware Fine-Grained Gap-Dependent Bounds for Online Reinforcement LearningDeep reinforcement learning methods → RL theory and sample complexity
Dynamic Routing as a New Dimension for Test-time Versatility of LLMsLLM reasoning → Test-time compute and efficient thinking
Agentic discovery of blood biomarker from distilled private health recordsLLM agents and tool use → Medical and scientific domain agents
Exact Fast Batch Simulation for Tabular Reinforcement LearningDeep reinforcement learning methods → RL theory and sample complexity
Lie Rarely, Lie Big: Stealthy Insider Attacks on LLM Robot TeamsLLM agents and tool use → Agent threats, guardrails and governance
Toward a Locally Deployable Agentic Co-Scientist: Small-Model Planning for Early-Stage Drug DiscoveryLLM agents and tool use → Medical and scientific domain agents
PyINE: A Framework for Scalable Elicitation and Oversight via Code ExecutionLLM reasoning → Reasoning for agents, planning and code
Verb-ICL: Rethinking In-Context Learning for Structured PredictionLLM reasoning → Reasoning benchmarks and evaluation
Knossos and Ariadne: Benchmarking and Learning Complete Diagram Topology Extraction with Vision-Language ModelsMultimodal LLMs / vision-language models → Chart understanding and anomaly detection
Towards Automatically Pruning Logging Code with Coding Agents: How Far Are We?LLM agents and tool use → Software engineering coding agents
Organising Trajectory Evidence for Language-Model Agent Assurance: Fragments, Methods, and the ResidualLLM agents and tool use → General LLM agent frameworks and benchmarks
Localized Operator Learning with Adaptive Partition-of-Unity Mixture-of-Expert NetworksMixture-of-experts → MoE theory and gating functions
Score-Calibrated Flow for Sampling from Unnormalized Densities with Applications to Generative Online Reinforcement LearningDiffusion models → Diffusion and flow policies for robotics +1
RAGStress: A controlled benchmark for evaluating retrieval-augmented generation under knowledge-base degradationRetrieval-augmented generation → RAG robustness and adaptive retrieval
When Debate Helps: Proposal Supply and Verification-Aware Readout in Multi-Agent ReasoningLLM agents and tool use → Social simulation and multi-agent debate
Steering Speech-Language Models: Training-Free Task Specialization via Contrastive Activation AdditionMechanistic interpretability → Activation steering methods
RoboIRS: Inference-Time Internal Representation Steering for Generalist Robot PoliciesWorld models → Embodied and robot world models +1
MASBench: Benchmarking LLM-based Multi-Agent Collaboration under Partial ObservabilityLLM agents and tool use → General LLM agent frameworks and benchmarks
PermVLA: Factorization Order as a Regularizer for VLA LearningVision-language-action models → VLA architectures and generalization
SEIS: Self-Evolving Inference SystemsSynthetic data and self-improvement for LLMs → Self-evolving LLM agents
Grounding Probes: Generator-Independent Hallucination Detection from Observer Model Hidden StatesRetrieval-augmented generation → Hallucination detection in RAG
Policy as Data: Replay-Based Policy Dual Averaging via Advantage RegressionDeep reinforcement learning methods → RL theory and sample complexity
Backward-Consistent Diffusion Sampling for Sparsely Observed PDE Inverse ProblemsDiffusion models → Flow matching, time series and science
PerturBot: Breaking Shortcut Priors in Vision-Language-Action Models with Perturbative TrainingVision-language-action models → Reasoning and chain-of-thought in VLA
ForeAct3D: Policy-Grounded Future World Modeling for VLA PoliciesWorld models → Embodied and robot world models +1
DiffGate: Difficulty-Gated Teacher Guidance for On-Policy DistillationLLM reasoning → Reinforcement learning for reasoning +1
SIFT: Robust Meta-Faithfulness Verification of Chain-of-Thought Reasoning Under Distribution ShiftLLM reasoning → Prompting and chain-of-thought reasoning
StegoMemory: Agentic Memory Acts as Covert Steganographic ChannelLLM agents and tool use → Agent threats, guardrails and governance
Frozen in a Frame: The Velocity Blind Spot in JEPA World ModelsWorld models → Latent predictive world models
Strong Helps Weak: Directional Cross-Modal Alignment Transfer in Multi-modal LLMsMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
From Probe Scores to Alarm Policies: Operational Validity of Activation Monitors for Language-Model AgentsLLM agents and tool use → Agent threats, guardrails and governance
Gradient-Free Sampling from Generative Models via Stochastic Bounded Extremum SeekingDiffusion models → Diffusion foundations and sampling
TRIZa: Guiding Agent Evolution from Pattern Recognition to Solution InventionLLM agents and tool use → Agent threats, guardrails and governance
EagleDepth: Efficient Fine-Grained Depth Estimation via Pixel Diffusion DecoderDiffusion models → 3D, point cloud and scene generation
DreamFormer: Dream Imitation with a Transformer World Model for Language-Conditioned Robotic ManipulationWorld models → Embodied and robot world models
Action-Consequence Alignment for Reliable Planning and Self-Improving in Latent World ModelsWorld models → Latent dynamics models and model-based RL
FASTER: Fast Adjoint Stochastic Transport for Endpoint Refinement in Reward-Guided Image EditingDiffusion models → Diffusion and flow policies for robotics +1
Quantifying Collusion Among Autonomous LLM Agents: A Statistical Analysis of the Collusion Wiki IncidentLLM agents and tool use → Agent threats, guardrails and governance
EvoCast: Reliable Autonomous Research Agents for Iterative Forecasting Architecture EvolutionLLM agents and tool use → Financial trading and forecasting agents
ManifoldCache: Training-Free Diffusion Acceleration via Constraint Manifold CachingDiffusion models → Flow matching, time series and science
EgoExo-Next:Benchmarking Vision-Language Models on Visual-Option Next-State and Cross-View ReasoningLLM reasoning → Multimodal reasoning +1
The Same Zero: Why Identical ASR Can Imply Different Guarantees in LLM-Agent SecurityLLM agents and tool use → Agent threats, guardrails and governance
Localization Lens for Improving Medical Vision-Language ModelsMultimodal LLMs / vision-language models → Medical vision-language models
Homogeneous Semantic Alignment and Hierarchical Expert Routing for Radiology Report GenerationMixture-of-experts → Medical and biomedical MoE
DreamTest: World-Model Surrogates for Search-Based Testing of Deep Reinforcement Learning AgentsWorld models → Latent dynamics models and model-based RL +1
Can LLM Agents Automate Reinforcement Learning for Text-to-Speech?LLM agents and tool use → Reinforcement learning for LLM agents
VCLMU: Mechanism-Centric Virtual Cell World Modeling for Perturbation ResponseWorld models → Medical and surgical world models
Reactivating Alignment: Defending LLMs from Jailbreaks via Intention-Aware Input-Output MatchingLLM safety and security → Jailbreaking
Trinity: Self-Evolving Vision-Language Models with a Self-VerifierLLM reasoning → Multimodal reasoning +2
Target-free Latent Safety AlignmentLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
CRAFT: An Agentic Spreadsheet Form Filling System with Template AwarenessLLM agents and tool use → Software engineering coding agents
Video2World: Benchmarking Coding Agents for Interactive World Modeling from Embodied VideosWorld models → Video-generation world models and simulators +1
UnAct: Gradient-Free Unlearning via Targeted Activation InterventionLLM safety and security → Machine unlearning
Measuring Effective Data Resolution in Guided Diffusion PosteriorsDiffusion models → Inverse problems and image restoration
CORE-RL: Confidence-Oriented Reliability Evaluation of Black-Box Reinforcement Learning PoliciesDeep reinforcement learning methods → Safe and constrained RL
Understanding and Mitigating Hallucination Escape in Tool-Using LLM AgentsLLM agents and tool use → Tool learning and function calling
AgenticTactileVLA: Contact-Guided Execution-Time Supervision for Generalizable Dexterous Manipulation without VLA RetrainingVision-language-action models → Tactile, force-aware and humanoid VLA
TrustMed-RL: Long-Horizon Reinforcement Learning for Evidence-Grounded Clinical DiagnosisDeep reinforcement learning methods → RL for clinical treatment and surgery
COPEX: Benchmarking LLM Robustness to Adversarial Context Across Model Context Protocol LayersLLM agents and tool use → Agent threats, guardrails and governance
Do Tool Calls Execute as Intended? Measuring and Repairing Intent-Execution Correspondence in LLM AgentsLLM agents and tool use → Software engineering coding agents
Human Behavior-Informed Crash Scenario Generation with Real-World Crash Priors for Autonomous Vehicle Safety EvaluationEnd-to-end autonomous driving → Adversarial scenarios and safety testing
Hierarchical Credit Assignment for RLVR on Fused Gromov-Wasserstein GeometryLLM reasoning → Reinforcement learning for reasoning
Bidirectional Preference Synthesis: Learning Prompt-Conditioned Preferences from Boundary FailuresLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Rethinking Long-Video Efficiency: A Joint Allocation Perspective on Frames, Pixels, and Front-End LatencyLLM reasoning → Multimodal reasoning +1
MOIRA: Mass-Oriented Indexing with Ragged Attention for Long-Context DecodingEfficient LLM inference and serving → KV cache optimization
Modeling Deletion Requests in Machine UnlearningLLM safety and security → Machine unlearning
Questioning the Questions: Sustaining Self-Evolution in Reasoning ModelsLLM reasoning → Test-time compute and efficient thinking
Language-Conditioned Token and Reasoning Efficiency in Large Language Models: A Paired Cross-Lingual Study ProtocolLLM reasoning → Reasoning benchmarks and evaluation
LMBuild: Evaluating LLM Agents for Generating Buildable and Functional StructuresLLM agents and tool use → Multimodal and embodied agents
First-Order Steering: Translating Weight Adaptation into Activation SteeringMechanistic interpretability → Activation steering methods
Dense Neuro-Symbolic Reasoning in a Unified Geometry StateLLM reasoning → Mathematical reasoning
Rethinking Self-Distillation for Multi-Teacher Capability MergingKnowledge distillation → On-policy and self-distillation for LLMs
Self-Reflection Fine-Tuning: Enhancing Agent Security against Prompt Injection Attacks from Failure ExperienceLLM agents and tool use → Agent threats, guardrails and governance +1
CADForge: Agentic Single-View CAD Reconstruction with Explicit Geometry ReasoningLLM agents and tool use → Multimodal and embodied agents

All 323