Papers

New on arXiv

Wed 16 Sep 2026

455 papers

When Keywords Drop but Classifiers Hold: Soft Refusals under KV Cache CompressionEfficient LLM inference and serving → KV cache optimization +1
KerColle: Unlocking Fine-Grained GPU Concurrency in Vision-Language-Action ModelsVision-language-action models → Efficient VLA inference and adaptation
AffordanceWAM: Affordance-Aware Joint World-Action Modeling for Robot ManipulationWorld models → Embodied and robot world models +1
Visual Graph Reasoning via Knowledge CompilationLLM reasoning → Multimodal reasoning
Proxifield: Decentralized Multi-Agent Communication through Semantic ProximityLLM agents and tool use → General LLM agent frameworks and benchmarks
Elastic Threshold Attention: Learned Contextual Sparsity for Long-Context DecodingEfficient LLM inference and serving → KV cache optimization
BirdsongChat: A Hybrid Multi-Agent Framework for Multimodal Embodied Behavior SimulationLLM agents and tool use → Multimodal and embodied agents
BI-Agent and BI-Bench: Towards Automating End-to-End Business IntelligenceLLM agents and tool use → Web, search, and research agents
Reputation as Community Memory for the Agentic WebLLM agents and tool use → Agentic recommender systems
Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-Time ScalingLLM reasoning → Test-time compute and efficient thinking
Efficiently Linking Unstructured Data for Multi-step ReasoningLLM reasoning → Knowledge- and tool-augmented reasoning
FASA: Feedback-Aware Sampling Adaptation for Efficient Diffusion-Based VLA ModelsVision-language-action models → Efficient VLA inference and adaptation
Safety Beyond the Interface: Detecting Harm via Latent States in Large Language ModelsLLM safety and security → Safety alignment, refusal and guardrails
Compositional Reasoning in Language Models under Reinforcement Learning Post-TrainingLLM reasoning → Reasoning benchmarks and evaluation
Beyond Private Training: The New Landscape of AI PrivacyRetrieval-augmented generation → Privacy and security of RAG
GLAMDRING: Gait Learning And Morphology co-Design via Reinforcement LearnING of CPGsDeep reinforcement learning methods → RL for robot locomotion and manipulation
Predict Before You Deploy: Offline Prediction of Quantization-Induced Task Degradation for World Action ModelsWorld models → Embodied and robot world models
Closed-World Resolution Against Tool Hallucination in LLM AgentsLLM agents and tool use → Medical and scientific domain agents
Less Is More: Graph-free Multimodal RAG via Multi-signal Late FusionRetrieval-augmented generation → Multimodal RAG
Improving Offline Goal-Conditioned Reinforcement Learning via Selective Reward StimulationOffline RL and imitation learning from data → Diffusion and flow policies for offline RL
WZPlanner: Safe End-to-End Path Planning for Autonomous Driving in Work ZonesEnd-to-end autonomous driving → VLMs for autonomous driving
MAGS: Multi-agent Auto-formalization Guarantees Safety for Agentic OutputsLLM agents and tool use → Agent threats, guardrails and governance
The Role of Fine-grained Harm Signals in LLM SafetyLLM safety and security → Safety alignment, refusal and guardrails
How to Guide Your Language FlowDiffusion models → Discrete and text diffusion
Can Vision-Language Models Judge Olympic Diving? From Reasoning to Scores in Zero-Shot Action Quality AssessmentLLM reasoning → Multimodal reasoning +1
A frontend-backend architecture for tool calls in full-duplex speech modelsLLM agents and tool use → General LLM agent frameworks and benchmarks
AUDITPLAN: Commit, Then Answer for Auditable Safety AlignmentLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
GAVEL: Graph World Models for Verified and Efficient Long-Horizon LLM Task PlanningWorld models → World models for language agents
Characterizing Web Search by Conversational LLM Agents: From Search Decisions and Strategies to Results and ResponsesLLM agents and tool use → Web, search, and research agents
Block Parallelism For Efficient Distributed Long-Context Diffusion Language Model TrainingDiffusion models → Discrete and text diffusion
PAPC: Platform Mediation for Privacy-Propagation Externalities in AI-Mediated WorkflowsLLM agents and tool use → Agent threats, guardrails and governance
MeshKV: A Network-on-Chip KV Cache Fabric for Scalable Transformer Decoding AcceleratorsEfficient LLM inference and serving → KV cache optimization
Position: It is Time to Virtualize Foundation Models with a Self-evolving Operating System LayerLLM agents and tool use → General LLM agent frameworks and benchmarks +1
EvoSherlock: Towards Agentic Lifelong Evolution for Unseen Long-Tailed Security-Critical Events in VideosLLM agents and tool use → Multimodal and embodied agents
ULOHA: An Underwater Bimanual Robot System for Robot LearningOffline RL and imitation learning from data → Robot manipulation imitation learning
Code-as-Auditor: Executable Compliance Reasoning via Regulation-to-CodeLLM reasoning → Reasoning for agents, planning and code
A Zeroth-Order Paradigm for LLM Preference AlignmentLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
PointZero: 3D Point Track Completion for Learning Transferable 3D DynamicsWorld models → Embodied and robot world models
In-Context Robot Learning with VLM AgentsLLM agents and tool use → Multimodal and embodied agents
Affora: A Design System for Agent-Friendly InterfacesLLM agents and tool use → GUI and mobile agents
rMuscle: Robotic Muscle Memory for Efficient Vision-Language-Action Model InferenceVision-language-action models → Efficient VLA inference and adaptation
Monitoring and Discovering Reward Hacking with Internal Representations during LLM EvaluationsLLM alignment / RLHF / preference optimisation → Reward hacking and misspecification +1
MUSE: Benchmarking Large Vision-Language Models on Multi-Modal Understanding in Situated EducationMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
RLLBC-Lib: An Educational Code Library for Reinforcement Learning and Learning-Based ControlDeep reinforcement learning methods → Deep RL algorithms and representations
Higher-order pruning of experts in mixture-of-experts language modelsLLM agents and tool use → General LLM agent frameworks and benchmarks +1
Beyond Outcomes: Dual-View Relational Learning for Efficient Agent BenchmarkingLLM agents and tool use → General LLM agent frameworks and benchmarks
Preventing Model Collapse: A Fisher-Rao Perspective on the Dynamics of Training with Synthetic DataSynthetic data and self-improvement for LLMs → Model collapse from synthetic data
ASLEval: Measuring Privacy Exposure Displacement in LLM Agent SessionsLLM agents and tool use → Agent threats, guardrails and governance
Decodable but Misrouted: Sparse Features Uncover a Readout Gap in Vision-Language Models for Harmful Meme DetectionMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Taming the Agentic RAN: Stability-Guaranteed Arbitration of Autonomous AI Agents in O-RANLLM agents and tool use → Agentic AI for wireless networks
Ask the Tool, Don't Guess: Agent Tool Calls Hold Their Progress, and the Serving System Should Read ItLLM agents and tool use → Software engineering coding agents +1
ReFigBench: Benchmarking Scientific Figure Reconstruction as Editable PowerPoint ArtifactsLLM agents and tool use → Multimodal and embodied agents
Infinite-Parameter LLMs: Generating and Adapting Weights from Live DataMixture-of-experts → MoE large language models
Compositional Policy Violations: When Step-Level Compliance Fails In Agentic AI WorkflowsLLM agents and tool use → Agent threats, guardrails and governance
ProgramDistill: From Interactive Web Apps to Verifiable Reference-Guided SWE TasksLLM agents and tool use → Software engineering coding agents
CERA-MoA: Co-Evolving Routing Mechanisms with Continually Learning LLM AgentsLLM agents and tool use → General LLM agent frameworks and benchmarks
Clueing up LLMs with Tool-Augmented Deductive ReasoningLLM reasoning → Reasoning for agents, planning and code
Beyond Truncation: Rethinking LLM Decoding as Ensemble PruningLLM reasoning → Test-time compute and efficient thinking
Calibrated Probabilistic Obstruction Reasoning with Vision-Language Models for Grasping in ClutterLLM reasoning → Spatial and embodied reasoning +1
Rethinking Critic Learning in PPO: Understanding and Mitigating Value FlatteningLLM reasoning → Reinforcement learning for reasoning
Generalist-Specialist Mixture-of-Experts for Rare Pathology Detection in Multimodal ImagingMixture-of-experts → Medical and biomedical MoE
WeaveRL: Weaving Reconstruction into Scene-Aware Fabrics for Perceptive Reinforcement LearningDeep reinforcement learning methods → RL for robot locomotion and manipulation
Voice of Reason: Reinforcement Learning for Spoken MathLLM reasoning → Reinforcement learning for reasoning +1
HBFlex: A Flexible Memory System for Bridging Fine-Grained LLM States and Coarse-Grained HBF Parallel ExecutionEfficient LLM inference and serving → Serving systems and request scheduling
CaMeLoT: CaMeL orchestrated with Temporal logic for static verification and livenessLLM agents and tool use → Agent threats, guardrails and governance
FIVE-VLA: Fast and EffectIVE Autonomous Driving with Recurrent Action MemoryVision-language-action models → VLA for autonomous driving +1
PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?LLM agents and tool use → Agent threats, guardrails and governance
Online Robust Reinforcement Learning Through Monte-Carlo PlanningDeep reinforcement learning methods → RL theory and sample complexity
Hypothesis-Driven Autonomous Materials Synthesis with Multimodal LLM AgentsLLM agents and tool use → Medical and scientific domain agents
Reasoning through Evolution: Automatic Meta-path Discovery for LLM-based Fake News DetectionLLM reasoning → Reasoning benchmarks and evaluation
Recursive Reasoning or Statistical Extrapolation? In-Context Learning in Multi-Agent Interdependent Decision-MakingLLM reasoning → Reasoning for agents, planning and code +1
Label-free steering: Compressing test-time reinforcement learning into bias-only subspacesDeep reinforcement learning methods → RL training of language and agent models
GroundingVLN: Reasoning and Acting with Grounding for Vision-Language NavigationLLM reasoning → Spatial and embodied reasoning +1
A Probe Shift Is Not a Fairness Fix: The Limits of Representation Steering in Speech ModelsMechanistic interpretability → SAE features and their applications
AeroWeaver: An Embodied-Agent Harness for Weaving Aerial Skills into Distributed, Adaptive Swarm ExecutionLLM agents and tool use → Multimodal and embodied agents
Beyond Routine Compliance: Cunning Data Cultivates Safety Vigilance in Large Language ModelsLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
DiT-Garment: Garment Dynamics with Diffusion TransformersDiffusion models → 3D, point cloud and scene generation
TAO-Force: Unifying Force-Aware Perception and Fast-Slow Control for Contact-Rich ManipulationVision-language-action models → Tactile, force-aware and humanoid VLA
Learning A Unified Template for Gait RecognitionDiffusion models → Video generation
Beyond Random Couplings: Contrastive Noise Alignment in Generative FlowsDiffusion models → Flow matching, time series and science
ActionPiece: Rethinking Action Tokenization for Autoregressive Vision-Language-Action ModelsVision-language-action models → Efficient VLA inference and adaptation
First Token Matters: Understanding Safety Collapse in Large Reasoning ModelsLLM reasoning → Reasoning for agents, planning and code +2
Spatially Adaptive Noise InjectionDiffusion models → Inverse problems and image restoration
CSWAM: Better Causal Semantic Representations for Out-of-Distribution Generalization in World Action ModelsWorld models → Embodied and robot world models
Disentangling Long-Term Memory via Latent Neuro-Symbolic ReasoningLLM reasoning → Reasoning for agents, planning and code
Collective Loss of Control in LLM Agent Systems: An Epidemic Account of Mutation, Contagion, and RecoveryLLM agents and tool use → Agent threats, guardrails and governance
The Mirage of Calibrated Confidence: Trajectory-Independence of Verbalized Confidence in Vision-Language ModelsMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
M-SQE: Multilingual Skill Quality Estimation for Enhancing Language Equality in Agentic Skill UseLLM agents and tool use → Agent skill learning and evolution
Risk-Aware World Modeling with Flow-Guided Occupancy Evolution for Selective Trajectory Planning in Automated DrivingWorld models → World models for autonomous driving +1
Multitask Reinforcement Learning for Assisting Choice Model SpecificationDeep reinforcement learning methods → RL training of language and agent models
Planning or Improvisation? Stress-Testing the Poetry Planning Site on Open Models and Open Cross-Layer TranscodersMechanistic interpretability → Steering and interpreting LLM reasoning
Exploring LLMs and RAG for Plausible and Explainable Material Prediction of Vehicle ComponentsRetrieval-augmented generation → Medical and biomedical RAG
WetRobo: A Reproducible Robot Kit for Coding Agents in Biological LaboratoriesLLM agents and tool use → Web, search, and research agents
Occluded Gait Recognition with Mixture of Experts: An Action Detection PerspectiveMixture-of-experts → MoE in reinforcement learning and robotics
StrucPhysVideo: Learning Physical Dynamics from Structured Captions and Robot ActionsWorld models → Latent predictive world models +1
Dependency-Aware Trajectory Refinement for Efficient Multi-Turn Agent Fine-TuningLLM agents and tool use → General LLM agent frameworks and benchmarks
The Verifiable Action Card: Trustworthy Human-in-the-Loop Control for Secure Autonomous AgentsLLM agents and tool use → Agent threats, guardrails and governance +1
DistAL: Distance-based Advantage Learning for VLA Fine-TuningVision-language-action models → Reinforcement learning and post-training +1
Decoupling Vision, Language, and Action for Efficient Multi-Task Robot PoliciesMultimodal LLMs / vision-language models → Embodied and 3D vision-language models +1
Market Signal Injection: Adversarial Context Manipulation of LLM Pricing AgentsLLM agents and tool use → Financial trading and forecasting agents

All 455