Papers

New on arXiv

Tue 25 Aug 2026

468 papers

Stale Does Not Mean Unsafe: Guard Precision for Tool-Using LLM Agents under Infrastructure State RacesLLM agents and tool use → Agent threats, guardrails and governance
BiGraph-Diffuse: A Bidirectional Diffusion Language Model with Graph-Structured Retrieval For Mental Health CounselingDiffusion models → Discrete and text diffusion
CataOPD: Catalytic On-Policy Distillation for Large Language Model ReasoningLLM reasoning → Prompting and chain-of-thought reasoning +1
AdaPilot: Towards Scene-Adaptive Policy Learning for Cross-Generator Text-to-Image Quality OptimizationDeep reinforcement learning methods → RL with diffusion and flow models
VIVAS: Vitalizing Visual Perception in VLM Pre-training via Vision-language Unified Autoregressive SupervisionMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive ParadigmMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
MIRAGE: How Conversation State Shapes Historical Evidence Use in Multimodal Personal AgentsMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning +1
Cognitively-Grounded On-Device Runtime Learning for Ground Robots in Unknown Physical EnvironmentsDeep reinforcement learning methods → Safe and constrained RL
ADeptS-Bench: Measuring the Trustworthiness of Computer Use Agents Across DevicesLLM agents and tool use → Agent threats, guardrails and governance
GameWAM: A World Action Model for Video GamesWorld models → Video-generation world models and simulators
Benchmarking AI Agents for Hardware Design Automation via MCP Tool CallingLLM agents and tool use → Software engineering coding agents
Agentic AI for operating scientific instruments for nanoscale characterizationLLM agents and tool use → Medical and scientific domain agents
Harness Engineering for Predictable Agentic Systems: An Empirical Study of Deterministic Execution ConstraintsLLM agents and tool use → Software engineering coding agents
Cost-Utility Alignment in LLM Agent Trajectories:Profiling,Attribution,Diagnosis,Adaptation,and EvaluationLLM agents and tool use → General LLM agent frameworks and benchmarks
Long-Term Behavioral Evaluation for Trusted Collaborator Selection via Bidirectional MambaState-space sequence models (S4, Mamba) → Video, motion and generative Mamba models
Output Dilution: Redundant but Fragile Representations in MoE ModelsMixture-of-experts → MoE theory and gating functions
Trust the Mass: Forced Weights in KV-Cache EvictionEfficient LLM inference and serving → KV cache optimization
Federation Is Nearly Free, Reasoning Is Not: Tradeoffs for AI Co-Scientists in Protein Characterization WorkflowsLLM reasoning → Clinical and biomedical applications
SpecMine: A Large-Scale Corpus of Spec-Driven Development ArtifactsLLM agents and tool use → Software engineering coding agents
MoPLEx: Estimating Plackett-Luce Mixture Models for Multi-Objective AlignmentLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Tunable Tool-Call Rates in LLM Agents via Representation SteeringLLM agents and tool use → Tool learning and function calling +1
Simulating Cognitive Smart Freight Corridors with Agent-Based Models and Reinforcement LearningDeep reinforcement learning methods → RL for energy, networks and operations
AudioLens: Multi-Perspective Speech Clustering with Reasoning Audio-Language ModelsLLM reasoning → Audio-language model reasoning
Model-Based Agentic Software EngineeringLLM agents and tool use → Software engineering coding agents
Bayesian Flow Networks for Offline Trajectory PlanningOffline RL and imitation learning from data → Diffusion and flow policies for offline RL +1
Belief Cascades Drive Persuasion in LLM Agent NetworksLLM agents and tool use → Social simulation and multi-agent debate
Drift Variation Autoencoder: Unifying Generation and Representation Learning through Conditional Posterior Flow MatchingDiffusion models → Flow matching, time series and science
SelfGraphRAG: Bridging the Supervision Gap in Graph-Based RAG with Synthetic QA GenerationRetrieval-augmented generation → Knowledge-graph and structured-knowledge RAG
Less can be More: Relieving RAG Bottlenecks via Evidence Frontloading and Pressure-Adaptive BudgetingRetrieval-augmented generation → Generation and context integration +1
Towards Reliable, Generalizable, and Specific In-Context Knowledge Editing via Multi-Objective Reinforcement LearningDeep reinforcement learning methods → RL training of language and agent models
Auto-Policy, not Auto-Skill: Compiled Agent Skills for the Physical WorldLLM agents and tool use → Agent threats, guardrails and governance
FLINT: Efficiently Leveraging High Bandwidth Flash for Capacity-Scalable LLM Inference AccelerationEfficient LLM inference and serving → Serving systems and request scheduling
LifePlanner: Evaluating LLM Agents for Geo-spatial Planning with Social Media DataLLM agents and tool use → Multimodal and embodied agents
Retrieve, Match, Escalate: Accurate and Scalable Product Linking with VLM-Distilled Cross-Encoders and Agentic VLMsLLM agents and tool use → Social simulation and multi-agent debate
Rollout-Decoded Reconstruction for Long-Horizon Prediction in Latent World ModelsWorld models → Latent predictive world models
Does Fine-Tuning Undo Activation Steering? Behavioural Recovery Without Weight-Edit ReversalMechanistic interpretability → Activation steering methods
D3-MOPD: Dynamic Domain ScheDuling for Efficient Multi-Teacher DistillationKnowledge distillation → On-policy and self-distillation for LLMs
Retrieved But Not Reliable: A Survey on Attacks, and Defenses in Retrieval-Augmented GenerationRetrieval-augmented generation → Privacy and security of RAG
Targeting the Attention Heads Behind Object Hallucination in LLaVAMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
GaussVLA: Geometry-Aware Spatial Reasoning for Vision-Language-Action ModelLLM reasoning → Spatial and embodied reasoning +1
Can We Read the Mind of an Audio LLM? A Verbalizable, Multilingual Middle-Layer WorkspaceLLM reasoning → Audio-language model reasoning +1
ToolMinimize: Auditing and Rewriting LLM Agent Tool Calls to Minimize Privacy ExposureLLM agents and tool use → Agent threats, guardrails and governance
AFDBench: A Reasoning-First AI Scientist for NationalWeather Service Forecast DiscussionsLLM reasoning → Time series reasoning
Do Robotic World Models Really Follow Actions? Diagnosing and Aligning Action-Conditioned Generation for Policy LearningWorld models → Embodied and robot world models
Latent Action as Intention Enables Efficient Future Imagination for World Action ModelsWorld models → Embodied and robot world models
Recursive Experiential-Working Memory Evolution for Long-Horizon Agent HarnessesLLM agents and tool use → Agent skill learning and evolution
SPO++: Stream-Aligned Policy Optimization for Asynchronous Agentic RLLLM agents and tool use → Reinforcement learning for LLM agents
Bellman Calibration for Marginalized Importance Weighting in Offline Reinforcement LearningOffline RL and imitation learning from data → Offline RL algorithms
LeFlow: Generative Latent Flow Planning for World ModelsWorld models → Latent predictive world models
BrowserForge: Scaling Web Episode via Parallel Browser SandboxesLLM agents and tool use → Web, search, and research agents
Strictly Causal Streaming Video Anomaly Detection with a Theoretically-Grounded State-Space CoreState-space sequence models (S4, Mamba) → Time series forecasting with SSMs
Structurally-bounded Agentic Graph Exploration for Evidence-Grounded Scholarly DeepSearchLLM agents and tool use → Web, search, and research agents
CAFE: Self-Improving Search Agents Need Co-Evolving FeedbackLLM alignment / RLHF / preference optimisation → Process reward models for reasoning
Right Diagnoses, Decorative Reasoning:A Perturbation Audit of Medical Chain-of-ThoughtLLM reasoning → Clinical and biomedical applications
Linear Probing Provides Robust and Efficient Detection of Machine-Generated TextMechanistic interpretability → Probing deception, bias and beliefs
StepGuard: Learning Step-Level Guardrails with Scalable Supervision and Safety-Utility BalancingLLM agents and tool use → Agent threats, guardrails and governance +1
Evidence Blindness in Direct Corpus Interaction: Persistent Navigation with AtlasNavRetrieval-augmented generation → Multimodal RAG +1
MoTE: Mixture of Task Experts for Multi-Task Video UnderstandingMixture-of-experts → Mixture of LoRA experts
IDeaL: Data-Free Multi-Teacher Distillation via Improved Dead LeavesKnowledge distillation → General knowledge distillation methods
RACE: Scalable Statistical Estimation of Functional Consistency in LLM NeuronsMechanistic interpretability → Circuits and mechanistic analysis
The RAT: A Unified Bayesian Model for RAG EvaluationRetrieval-augmented generation → RAG robustness and adaptive retrieval
SkillForge: Evolving Verifiable Skills for Reinforcement Learning AgentsLLM agents and tool use → Agent skill learning and evolution +1
One-Shot Learning from Demonstration of Contact-Rich Robotic Manipulation by Identifying Physical InteractionsOffline RL and imitation learning from data → Robot learning from demonstration
Metan: Recursive Self-Improvement through Emergent DepthLLM agents and tool use → General LLM agent frameworks and benchmarks +1
Learning to Prefer Reliably: Error-Augmented Emotion Preference Optimization with Calibrated FusionMultimodal LLMs / vision-language models → Multimodal emotion understanding +1
GaussianWAM: Distilling Geometry and Semantics from 3D Gaussian Fields into World-Action ModelsWorld models → Embodied and robot world models
On-policy Distillation with Verifiable RewardKnowledge distillation → On-policy and self-distillation for LLMs
Confident at the moment of action: belief miscalibration in LLM play under hidden informationLLM agents and tool use → Reinforcement learning for LLM agents
Game2World Engine: Unlocking In-the-Wild Gameplay Videos for World Model TrainingWorld models → Video-generation world models and simulators
EviGraph: Towards Verifiable Evidence Construction for Information-Seeking AgentsLLM agents and tool use → Web, search, and research agents
Parason: Revealing Subtask and Trial Parallelism in LLM ReasoningLLM reasoning → Test-time compute and efficient thinking
Simthesizer: An Agent-Driven Simulation Framework for LLM Serving SystemsLLM agents and tool use → Software engineering coding agents
On-Policy Self-Distillation in Diffusion ModelsDiffusion models → Diffusion and flow policies for robotics +1
A Literate Programming Environment for Human and Machine AgentsLLM agents and tool use → Software engineering coding agents
Gripper-aware Vision Language Action ModelsVision-language-action models → Generalist VLA models and benchmarks
Is Discrete Difficulty Sufficient? Leveraging Continuous Difficulty for Efficient Self-Consistency in LLMsLLM reasoning → Test-time compute and efficient thinking
IAPO: Influence-Aware Policy Optimization for Credit Assignment in Multi-Turn Service AgentsLLM agents and tool use → Reinforcement learning for LLM agents
PhysMLLMs: Spatial Priors for Unified Referring Segmentation and Grounded Reasoning of Images and VideosLLM reasoning → Multimodal reasoning +1
Joint Optimization of Tool Creation and Use for Large Language Model AgentsLLM agents and tool use → Reinforcement learning for LLM agents
EviDx: Evidence-Aware Active Diagnosis with Scaffolded LLM AgentsLLM agents and tool use → Medical and scientific domain agents
When "Must" Becomes "Maybe": Constraint Weakening in LLM Agent WorkflowsLLM agents and tool use → General LLM agent frameworks and benchmarks
SeisMamba: Low-Latency Single-Station Seismic Magnitude Estimation for Spatially Distributed Earthquake Early WarningState-space sequence models (S4, Mamba) → Time series forecasting with SSMs
Neurosymbolic Alignment for Physiologically-Safe Clinical Language ModelsLLM alignment / RLHF / preference optimisation → Medical LLM alignment
RoG-DAgger: Rollout-Guided Post-Training for End-to-End DrivingEnd-to-end autonomous driving → Imitation-learning end-to-end planners +1
PeakBench: Benchmarking Resource-Aware Tool Invocation in LLM AgentsLLM agents and tool use → Software engineering coding agents
NeuralParker: A Reinforcement Learning Planner for Irregular Parking EnvironmentsDeep reinforcement learning methods → RL for autonomous driving and traffic
Beyond Static Interpretability: Anticipating Post-SFT Mechanisms from Pre-SFT Parameters for Better TuningMechanistic interpretability → Circuits and mechanistic analysis
WarpSAC: Towards the Pinnacle of Scalable Off-policy RL by Rethinking Exploration and ExploitationDeep reinforcement learning methods → Deep RL algorithms and representations
Ockhamareto: Pareto-Gated Segment-Level Credit Assignment for Concise Unit-Test Generation with Reinforcement LearningLLM reasoning → Reinforcement learning for reasoning +1
Reinforcement Learning-Guided Evolutionary Policy Optimization for Preference-Adjustable Heterogeneous Agile Earth Observation Satellite SchedulingDeep reinforcement learning methods → RL for robot navigation and UAV control
HMGCLIP: Heterogeneous Multi-Granularity Contrastive Learning for E-commerce Representation LearningMultimodal LLMs / vision-language models → Document understanding and multimodal RAG
DoublesEval: Diagnosing Multi-Agent Tactical Reasoning in Vision-Language Models via Professional Doubles BadmintonLLM reasoning → Multimodal reasoning +2
Vision Language Model Fusion for Explainable Face RecognitionMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
ResiSpec: Enhancing Multi-Candidate Speculative Sampling via Residual Distribution ShapingEfficient LLM inference and serving → Speculative decoding
Variance-Guided Spatial Attention Fusion for Robust End-to-End Driving under Asymmetric Sensor DegradationEnd-to-end autonomous driving → End-to-end system architecture
Rethinking Semantic Alignment in LLM-Enhanced Collaborative Filtering: A Spectral Decoupling ApproachLLM reasoning → LLM reasoning for recommendation
Adaptive Influence Graphs for Failure Attribution in Multi-Agent SystemsLLM agents and tool use → Software engineering coding agents
The Handoff Tax: Continuing Non-Native Trajectories in LLM AgentsLLM agents and tool use → Software engineering coding agents
Not All Tokens Are Equal: Region-Aware Consistency Repair of Backdoors in MLLMsLLM safety and security → Backdoor and data poisoning attacks
FARCA: Fact-Aligned Reliability-Aware Credit Assignment for Reinforcement Learning with Factual SupervisionDeep reinforcement learning methods → RL training of language and agent models

All 468