Papers

New on arXiv

Wed 2 Sep 2026

453 papers

Integrating Fairness and Explainability in a Multiple Instance Reinforcement Learning SystemDeep reinforcement learning methods → RL for games and combinatorial optimization
Silent Sabotage: Internal State Triggered Backdoor Attacks on LLM-Powered Robotic SystemsLLM safety and security → Prompt injection and agent security
EAVer: Long-Form Factuality Verification as an End-to-End Agentic PolicyLLM agents and tool use → Web, search, and research agents
Can Coding Agents Reproduce Official Statistics? Metadata, Retry Budget and the Limits of Execution Feedback in a Controlled Eurostat BenchmarkLLM agents and tool use → Web, search, and research agents
Team DArgk at the 2026 ELOQUENT lab for evaluating generative language model quality: Residuals of Humanity: AI Detection Evasion via GRPO Fine-TuningLLM reasoning → Reinforcement learning for reasoning
Knowing, and Saying It Only When Asked: LLM Endognostics and the Schizognosis of Minerva-7BLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Toollery: Scaling LLM Agents to Thousands of Skills and ToolsLLM agents and tool use → Tool learning and function calling
On Mitigation of Subliminal Learning in Large Language ModelsKnowledge distillation → On-policy and self-distillation for LLMs
Clinical Reasoning Under a Partially Observed Objective in Cone Beam CT Report GenerationLLM reasoning → Clinical and biomedical applications
Self-Evolving AI for Humanoids: Mechanisms, Safety, and Evaluation of Post-Deployment Self-ImprovementWorld models → World models for navigation +1
Harnessing human expertise for high-precision robotic assembly in industrialized construction: A sample-efficient installer-in-the-loop interactive reinforcement learning frameworkOffline RL and imitation learning from data → Real-world robot offline RL +1
ShieldVLA: Feasibility-Aware Safety Alignment for Vision-Language-Action ModelsVision-language-action models → Safety and adversarial attacks +2
Knowing When Not to Answer: Abstention and Refusal Reasoning in Vision--Language ModelsLLM reasoning → Clinical and biomedical applications
Dual-Latent Memory Routing for Vision-Language ReasoningLLM reasoning → Multimodal reasoning +1
Beyond the Verdict: Evidence-Aligned Evaluation of Visual Prompt-Injection GuardrailsMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness +1
SimpleMemVLA: A Simple but Effective Native-Video Memory for Vision-Language-Action ModelsVision-language-action models → Reasoning and chain-of-thought in VLA
FailSAE: Towards Interpretable Failure Prediction for Vision-Language Models via Sparse AutoencodersMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning +1
Reviewer Capability Governs Rejection Targeting, Not Repair Skill: Evidence from LLM Execute-Review-Revise PipelinesLLM agents and tool use → General LLM agent frameworks and benchmarks
Spectral-Target Physical Latent Structuring for JEPA-Style World ModelsWorld models → Latent predictive world models
Quality Recovery for Quantized KV Caches via Low-Rank Attention AdaptationEfficient LLM inference and serving → KV cache quantization
MemoryLACE: Memory Lifecycle-Aware Consolidation and Evidence RetrievalLLM agents and tool use → Agent memory systems
Where Reliability Lives: Experimental Localisation of Behavioural Properties in an Agent SystemLLM agents and tool use → General LLM agent frameworks and benchmarks
Jina-OCR-v1: Efficient Document Parsing with Speculative Decoding and Dense Verifiable RewardsEfficient LLM inference and serving → Visual token pruning in multimodal models
Frontier LLMs are effective batch optimizers: Assessing reasoning models in continuous and discrete settingsLLM reasoning → Test-time compute and efficient thinking
Who Speaks for the Pruned? Visual Token Pruning as Coverage OptimizationMultimodal LLMs / vision-language models → Visual token pruning and efficiency +1
VeriPhy: Agentic Physical Reasoning for World Model Evaluation and RefinementWorld models → Latent predictive world models +1
BASP: Communication-Efficient Batch-Aware Sequence Parallelism for LLM TrainingLLM reasoning → Test-time compute and efficient thinking
Routing Is Not Enough: Diagnosing Intra-Adapter Subspace Contention in MoE+LoRA Fine-TuningMixture-of-experts → Expert routing strategies
Sensing Which Modality Matters: Evidence-Gated Regularization for Robust VLA PoliciesVision-language-action models → Generalist VLA models and benchmarks
SLIDEFORGE: An LLM Agent for Controllable Editing of Slides as Structured ArtifactsLLM agents and tool use → Multimodal and embodied agents
The Gradient Does Not See Rank: Rank-Indifference in Matrix-CODI on ProsQALLM reasoning → Prompting and chain-of-thought reasoning
Solving the Needle-in-a-Haystack Problem in Mammography Vision-Language Model with Differentiable Subset SamplingMultimodal LLMs / vision-language models → Medical vision-language models
GPU-Accelerated Astrodynamics World Models for Spacecraft Rendezvous and Proximity OperationsWorld models → World models for navigation
You Can't Escape Your Own Activations : Evaluation Awareness and Multi-Agent MonitoringLLM agents and tool use → Agent threats, guardrails and governance
Requirements After the First Edit: Mining Late Requirement Emergence and Rework in Real-World Coding-Agent SessionsLLM agents and tool use → Software engineering coding agents
ObserverBench: Testing Mechanistic Estimates for Intervention and ControlMechanistic interpretability → Circuits and mechanistic analysis
Verify Before You Distill: Prompt-Level Teacher Gating for On-Policy DistillationKnowledge distillation → On-policy and self-distillation for LLMs
Tail-Likelihood Reinforcement LearningDeep reinforcement learning methods → RL theory and sample complexity
Privacy-Preserving Topology-Guided Safety for LLM-Based Multi-Agent Systems via Federated Graph LearningLLM agents and tool use → Agent threats, guardrails and governance
LexIssue: Benchmarking Legal Issue Identification in Chinese Civil LitigationRetrieval-augmented generation → Legal and financial RAG
SolarWM: Open Data and Scalable Training for Long-Horizon Video World ModelsWorld models → Video-generation world models and simulators
Discriminative World Models for Web AgentsWorld models → World models for language agents +3
Post-Training Language Models for Gold-Medal Performance in Coding CompetitionsLLM reasoning → Reasoning for agents, planning and code
Cliff: Learning Process Rewards from the First MistakeLLM reasoning → Reinforcement learning for reasoning +2
GDB-Reward: From Evaluation Metrics to Training Rewards for Graphic DesignDeep reinforcement learning methods → RL with diffusion and flow models
Do Better Imagined Rollouts Mean Better Robot Control? A Controlled Study of World-Model Evaluation Under FeedbackWorld models → Embodied and robot world models
Large Language Models (LLMs) for Telecom Root Cause Analysis (RCA): A Structured Reasoning Framework for Evidence-Grounded DiagnosisLLM reasoning → Reasoning for agents, planning and code
SafeEvolve: Harness-Policy Co-Evolution from Agent Experience for Safety AlignmentLLM agents and tool use → Agent threats, guardrails and governance +2
EarlyEval: Cheaper Agent Evaluation via Early Outcome PredictionLLM agents and tool use → General LLM agent frameworks and benchmarks
ShallowStream: Index Shallow then Answer Deep for Streaming Video UnderstandingEfficient LLM inference and serving → Visual token pruning in multimodal models
CodePoisonRAG: Knowledge Poisoning Attacks on Retrieval-Augmented Code GenerationRetrieval-augmented generation → Privacy and security of RAG
Measurement-Driven Sub-Network Selection for On-Premise Retrieval-Augmented Factory AgentsRetrieval-augmented generation → Agentic RAG, agents and memory
Untangling the Mechanisms of Misleading Context in Medical Question AnsweringLLM reasoning → Clinical and biomedical applications
Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM SystemsLLM agents and tool use → General LLM agent frameworks and benchmarks
Balancing Frequencies and Pixels in Flow MatchingDiffusion models → Flow matching, time series and science
Incremental Pooled LLM Evaluation for Cost-Effective Retrieval Model SelectionRetrieval-augmented generation → RAG pipeline design and evaluation
Language Models Can Control Their Own AttentionEfficient LLM inference and serving → KV cache optimization
RVSD: Retrieval Vision Sparse Decoding for Mitigating Visual Hallucinations in Large Vision-Language ModelsMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
BuildOcc: A Large Language Model Occupant Agent Platform for Building Energy ResearchLLM agents and tool use → Financial trading and forecasting agents
Trace as State: Reasoning Traces as Conditional States for Long-Context TransformersLLM reasoning → Test-time compute and efficient thinking
ACLE-MCP: Attested Capability Leases for Execution-Time Trust in Remote LLM Tool UseLLM agents and tool use → Agent threats, guardrails and governance
From Proxy Learning to Driving Decisions: A Transfer-Based Framework for Evaluating Future-Aware Autonomous Driving PlannersWorld models → World models for autonomous driving +1
Eliciting ESG Preferences for Reinforcement Learning-Based Portfolio OptimizationLLM alignment / RLHF / preference optimisation → LLM preference optimization methods +1
Characterizing Text Branch Sensitivity in Medical Vision-Language Segmentation via Evidence DecouplingMultimodal LLMs / vision-language models → Medical vision-language models
PrimSynth: An Agentic Approach to Discover, Validate, and Synthesize Exploit Primitives for Linux Kernel VulnerabilitiesLLM agents and tool use → Agent threats, guardrails and governance
Latent Cluster Analysis for Vision-Language-Action ModelsVision-language-action models → Generalist VLA models and benchmarks
Competitive Market Behavior of LLMsLLM agents and tool use → Financial trading and forecasting agents
Deeply Interleaved Text-Image Contexts for Multimodal LLMs AssessmentMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
MARS: What Retrieval Signals Are Hidden in Multimodal Large Language Models for Text-Video Retrieval?Multimodal LLMs / vision-language models → Video understanding with MLLMs
A Finger on the Scale: Covert Policy Steering through Agentic SkillsLLM agents and tool use → Agent threats, guardrails and governance
RGB-to-IR image translation for infrared vehicle detection in unseen UAV domainsDiffusion models → Image generation, text-to-image and editing
Learn from Whoever Is Right: Answer-Verified Multi-Teacher Distillation for Multi-Domain LLMsKnowledge distillation → On-policy and self-distillation for LLMs
ZETA: A Controlled Study of Zero-Shot Cross-Embodiment VLA Transfer for Tabletop ManipulationVision-language-action models → Human-video and latent-action pretraining
LightBridge: Feed-Forward Generative Relighting for 3D Gaussian SplattingDiffusion models → 3D, point cloud and scene generation
World-Model-Augmented Visual Locomotion for Humanoids on Foothold-Constrained TerrainWorld models → Humanoid and legged locomotion world models
A Comparative Study of Graph Representations for GNN-Based Power Grid Control in L2RPNDeep reinforcement learning methods → RL for energy, networks and operations
Spatially Aware World Action Model via Geometric Latent DiffusionWorld models → Embodied and robot world models +1
AceSpec: An Asymmetric Edge-Cloud Collaborative Framework for Communication-Efficient LLM InferenceEfficient LLM inference and serving → Early exit and edge inference
Beauty is in the AI of the beholder: MLLMs systematically overrate facial attractivenessMultimodal LLMs / vision-language models → Social bias and fairness in VLMs
Rethinking the Teacher-Student Framework for Test-Time AdaptationKnowledge distillation → General knowledge distillation methods
ViSAR: Training-Free Adaptive-k Retrieval for Visual Document Question AnsweringMultimodal LLMs / vision-language models → Document understanding and multimodal RAG +1
VIPS: Vehicle-Infrastructure Cooperative Planning Benchmark via Pseudo-SimulationEnd-to-end autonomous driving → Cooperative and V2X driving
CivBench: A Long-Horizon Benchmark for Tool-Mediated Agents in Civilization VILLM agents and tool use → Reinforcement learning for LLM agents
When Decodability Is Not Enough: Logical Validity Representations, Behavioral Dissociation, and Causal Tests in Language ModelsLLM reasoning → Reasoning benchmarks and evaluation
Coverage, Not Targeting: A Structural Regime in Multi-Turn Agent Credit AssignmentLLM agents and tool use → Reinforcement learning for LLM agents
Before the Script, Set the Stage: How Worldview Simulation Amplifies Psychologically Grounded Persuasion in Multi-Turn JailbreakingLLM safety and security → Jailbreaking
Evidence for Shared Routing Geometry and Dynamics in Sparse Mixture-of-ExpertsMixture-of-experts → Expert routing strategies
CA-OPD: Confidence-Aware On-Policy Distillation for Structured Visual PredictionMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning +1
Improving Health Literacy through Lay Summarization of Radiological Reports: An Evaluation of BioNER and Retrieval-Augmented GenerationRetrieval-augmented generation → Medical and biomedical RAG
CAPTCHAs in the Agentic Era: Solvers That Learn from Every EncounterLLM agents and tool use → Multimodal and embodied agents
Diagnosing with Insights: Structured Analysis of Agent Failures via Behavioral AbstractionsLLM agents and tool use → Software engineering coding agents
NE-R1: Enhancing Named Entity Recognition Model via Reinforcement LearningRetrieval-augmented generation → Generation and context integration
TempoGround: State-Aware Streaming Visual Grounding with Vision-Language ModelsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Structured-Prior-Guided Diffusion Inpainting with Physical Consistency for Traffic Sign AugmentationDiffusion models → 3D, point cloud and scene generation
Towards Zero-Shot Transfer Across Embodiments For Driving VLAsLLM reasoning → Spatial and embodied reasoning +2
AGI Maze Prediction Datasets: A Compact Benchmark for Learning World Dynamics with TransformersWorld models → Latent predictive world models
SALA: Semantic-Aware Logical Alignment for Complex Reasoning in In-Context LearningLLM reasoning → Test-time compute and efficient thinking
DiffIE: Diffusion-based Open Information ExtractionDiffusion models → Discrete and text diffusion
Efficient GUI Agents: A Systems Survey of Observation, Memory, Action, and Runtime OptimizationLLM agents and tool use → GUI and mobile agents
Improving Evaluation Realism with Inference-Time Compute and Deployment ScaffoldsLLM reasoning → Reasoning for agents, planning and code

All 453