Papers

New on arXiv

Wed 29 Jul 2026

439 papers

Improving Online Reinforcement Learning via Bidirectional Behavior Prior DistillationDeep reinforcement learning methods → Deep RL algorithms and representations
CleanVideo: Adaptive Concept Erasure for Text-to-Video Diffusion ModelsDiffusion models → Robustness, privacy and adversarial aspects
Disentangling Algorithmic Bias from Archival Artifacts: A Controlled Audit of Vision-Language Model Valuation in Metropolitan Museum ArchivesMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
PRISM: An Agentic Multi-Model Architecture for Proactive Safety in Autonomous Transportation SystemsLLM agents and tool use → Agent threats, guardrails and governance
Toward a social psychology of AI: language-model agents reproduce human-like minimal-group biasLLM agents and tool use → Social simulation and multi-agent debate
Preference Elicitation for Policy Optimization and Application to Aligning Heart Transplantation with Human ValuesLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
DUET: Dual-Teacher On-Policy Distillation via Same-Weight Disagreement for Prohibition ComplianceLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +2
Training and Evaluating Ethical Reinforcement Learning Agents on Per-Episode DistributionsDeep reinforcement learning methods → RL training of language and agent models
When LLM Agents Negotiate: Private Information and Dynamic Bargaining in Supply ChainsLLM agents and tool use → Social simulation and multi-agent debate
An evolutionary model of animats with VLM-based subjective evaluationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Beyond Accuracy: Auditing Spatial Provenance in Visual Token Pruning for OCR-Critical MLLM InferenceEfficient LLM inference and serving → Visual token pruning in multimodal models
Which Modality Decides? Counterfactual Modality Attribution for Multimodal LLMsMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
SafeBuild-Bench: A Temporal-Robust Construction Safety Benchmark with Graph-Enhanced Data MiningLLM safety and security → Safety alignment, refusal and guardrails
TELLER: Dual-Path Iterative Preference Optimization for Table Entity LinkingLLM reasoning → Knowledge- and tool-augmented reasoning +1
Reasoning in Real World Clinical Care: Why Large Language Models Are Not Yet Safe for Autonomous Clinical Decision SupportLLM reasoning → Clinical and biomedical applications
Latent-Kernel Discrete Flow Maps for Few-Step GenerationDiffusion models → Discrete and text diffusion
ThreatForest: Multi-Agent Attack Tree Generation with Pluggable TTP Framework MappingLLM agents and tool use → Agent threats, guardrails and governance
Hierarchical Reranking for Scalable Financial RAG SystemRetrieval-augmented generation → Legal and financial RAG
Automated Transcript Analysis for Detecting Flaws in Agentic BenchmarksLLM agents and tool use → General LLM agent frameworks and benchmarks
Belief Coevolution in a Social Network of Generalist and Specialist Large Language ModelsLLM agents and tool use → Social simulation and multi-agent debate
Failure Detection for Surgical Robot Imitation Policies via Flow-Matching World ModelingWorld models → Medical and surgical world models +2
Models for minimalist RAG: B1ade 335M Embedding and 1B Parameter Small Language ModelsRetrieval-augmented generation → Generation and context integration
SkillSmith: Learning to Compose Parametric Skills and Textual KnowledgeLLM agents and tool use → Agent skill learning and evolution
Skill Use or Skill Theater? Evaluating the Reasoning Backroom in Skill-Augmented Language AgentsLLM reasoning → Reasoning for agents, planning and code +1
Leveraging Trajectory Graphs for Pre-Execution Error Diagnosis in Agentic LLM SystemsLLM agents and tool use → Software engineering coding agents
SE(3)-MeanFlow: Few-Step Protein Backbone Generation on Lie GroupsDiffusion models → Flow matching, time series and science
Auditing Emergent LLM-Agent Collaboration through Cooperation-Obligation CouplingLLM agents and tool use → Agent threats, guardrails and governance
Bridging Inference-Time Scaling and Episodic Memory with Action-Centric GraphsLLM reasoning → Reasoning for agents, planning and code
SWE-NFI: Studying and Benchmarking Coding Agents for Non-Functional ImprovementsLLM agents and tool use → Software engineering coding agents
Beyond the Bidirectional Promise: Re-evaluating the Robustness of Diffusion Language ModelsDiffusion models → Discrete and text diffusion
Same Facts, Different Diagnosis: Measuring and Mitigating Narrative Anchoring in Clinical Language ModelsLLM reasoning → Clinical and biomedical applications
VideoCoCo: Code-as-CoT for Physically-Consistent Video Generation via an Agentic Dual-Engine SystemLLM agents and tool use → Multimodal and embodied agents
HSS-Synth: Humanities and Social Sciences Data Synthesis for LLMsSynthetic data and self-improvement for LLMs → LLM synthetic data generation
From Backlog Items to Security Guidance: Towards Continuous Security ComplianceRetrieval-augmented generation → Privacy and security of RAG
RoguePrompt: Dual-Layer Encoding for Self-Reconstruction to Circumvent LLM ModerationLLM safety and security → Jailbreaking
BridgeAlign: Bridging Preference Alignment for Humanities and Social SciencesLLM alignment / RLHF / preference optimisation → LLM preference optimization methods +1
SkillMentor: LLM Agent Self-Evolution via Learning Blind-Spot DiagnosisLLM agents and tool use → Agent skill learning and evolution
Shared Semantic Codebook Distillation for Unpaired Cross-Modal Medical ClassificationKnowledge distillation → Distillation for medical imaging
LayerRAG-Bench: A Cross-Layer Reliability Benchmark for Agentic Retrieval-Augmented GenerationRetrieval-augmented generation → Agentic RAG, agents and memory +1
Position, Not Provenance: Separating Reasoning Mediation from Sycophancy in Medical Vision-Language ModelsLLM reasoning → Clinical and biomedical applications +1
AgentS4D: Benchmarking Runtime Risks across the Execution Lifecycle of LLM-Based Workspace AgentsLLM agents and tool use → Agent threats, guardrails and governance
TIER-MoE: Trust-Informed Expert Routing via Conditional Modality Risk for Multimodal Fusion in Biomedical ClassificationMixture-of-experts → Medical and biomedical MoE
Open Security Benchmark: Towards Autonomous Enterprise Cyber DefenseLLM agents and tool use → Agent threats, guardrails and governance
The Kinetics of Training: A Driven-Nucleation Rate Law for Emergence, Plasticity Loss, and Circuit Control in Language ModelsMechanistic interpretability → Knowledge and representations inside LLMs
Flat Score, Amplified Failures: How the Error Budget Masks Damage in Quantized LLM AgentsLLM agents and tool use → Tool learning and function calling
Beyond KV Reconstruction: Functional Reconstruction for MLA Draft Models in Speculative DecodingEfficient LLM inference and serving → Speculative decoding
FAVA: Formal Authorization for Verified Agents with Evidence-Backed Permission GraphsLLM reasoning → Reasoning for agents, planning and code +1
It's Not Just More Demos: Counterfactual Action Sensitivity Coverage for Data-Efficient Robust Robot ImitationOffline RL and imitation learning from data → Robot manipulation imitation learning
TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAMVision-language-action models → Efficient VLA inference and adaptation
Do You Really Need to Pretrain Q-Functions for Online RL Fine-Tuning?Deep reinforcement learning methods → Deep RL algorithms and representations
Mental World ModelingWorld models → World models for language agents
HumanCLAW: Can Vision-Language Models Act Through a Body?Multimodal LLMs / vision-language models → Embodied and 3D vision-language models
SpecFirst: Behavioral Specification Elicitation as a First-Class Step in Agent-Based Program Synthesis from ScratchLLM agents and tool use → Software engineering coding agents
OmegaUse-OfficeVal: Benchmarking LLM Agents on Long-Horizon Office-Suite Tasks with Economic GroundingLLM agents and tool use → General LLM agent frameworks and benchmarks
Anatomy Contextualized Adaptation of CT Foundation ModelsMultimodal LLMs / vision-language models → Medical vision-language models
MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program SynthesisLLM agents and tool use → Software engineering coding agents
Explainable and Resource-Efficient Spatial Reasoning in Multimodal LLMs for Decision-Critical ApplicationsLLM reasoning → Spatial and embodied reasoning +1
DLAM: Distributional Latent Actions with Temporal ConstraintsVision-language-action models → Human-video and latent-action pretraining
KAMR: Grounding Generation via Knowledge-Aligned Multi-hop RetrievalRetrieval-augmented generation → Knowledge-graph and structured-knowledge RAG
Veritas++: Value-aware On-Policy Distillation for Perception-Enhanced AIGI DetectionKnowledge distillation → Multimodal distillation
FreqForcing: Autoregressive Long Video Generation via Spectral Self-AnchoringDiffusion models → Video generation
InferScale: GPU-Native KV Injection for Personalized LLM ServingEfficient LLM inference and serving → KV cache optimization
Scores Are Not Decisions: Cost-Aware Stopping for Tool Acquisition in LLM AgentsLLM agents and tool use → Tool learning and function calling
On-Policy Distillation for LLM Safety: A Routing Approach to Template-Robust RealignmentLLM safety and security → Safety alignment, refusal and guardrails +1
MemSecBench: Tracking Agent Memory Poisoning from Persistence to Consequence and RepairLLM agents and tool use → Agent threats, guardrails and governance
Visual Credit Audit for Multimodal Spatial ReasoningLLM reasoning → Spatial and embodied reasoning
Setoka: A Benchmark for Hierarchical User Understanding in Personalized Agents over Heterogeneous DataLLM agents and tool use → Agent memory systems
CoCaRS: Correlation Calibration-Based Redundancy Suppression for Heterogeneous Knowledge DistillationKnowledge distillation → General knowledge distillation methods
Mitigating Compounding Error via Video Representation RegularizationWorld models → Video-generation world models and simulators +1
What Can Latent World Models Know? Physical Parameter Identifiability in Multimodal Predictive RepresentationsWorld models → Latent predictive world models
AgentSnare: Learning to Delay, Divert, and Defuse Autonomous Penetration AgentsLLM agents and tool use → Agent threats, guardrails and governance
RL2-VLA: Adaptive RL Latent Compositional Steering with Test-Time Scaling for Vision-Language-Action ModelsLLM reasoning → Spatial and embodied reasoning +2
TREK: A Travel Reasoning and Evaluation Kit for LLM Agents in Complex Trip PlanningLLM reasoning → Reasoning for agents, planning and code +1
Credit Cards, Confusion, Computation, and Consequences: What Can We Uncover About Language Model Reasoning?LLM reasoning → Reasoning for agents, planning and code
Progressive Multimodal Alignment for Continual Instruction TuningMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
VITAL-RAG: Invariance Race for Context Allocation in Coding AgentsLLM agents and tool use → Software engineering coding agents
Temporally Centered SIGReg Improves LeWorldModel Representations for Robot Policy LearningWorld models → Latent predictive world models
Two Calls Beat Five Agents: Evaluating Multi-Agent Pipelines Against Self-Refinement for Local Language ModelsLLM agents and tool use → General LLM agent frameworks and benchmarks
Prior Directions: Why GUI Grounding Gets Locked in the PastMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
CinemaTraj: Composing Atomic Camera Trajectories for 3D Scenes with LLM AgentsLLM agents and tool use → Multimodal and embodied agents
Dual-Path LLM Reasoning for Multimodal Few-Shot Knowledge Graph CompletionLLM reasoning → Knowledge- and tool-augmented reasoning
DIRECT: Direct Decoding for Efficient and Aligned Sequence Labeling with Large Language ModelsLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Hearsay: Vision-Language Medical Diagnoses Without an ImageMultimodal LLMs / vision-language models → Medical vision-language models
SCALPEL: Semantic Cross-modal Alignment via LLM-Powered Encoder Learning for Medical Vision-Language RepresentationMultimodal LLMs / vision-language models → Medical vision-language models
SERPO: Self-Evolving Rubric Policy Optimization for Open-Ended Test-Time Reinforcement LearningSynthetic data and self-improvement for LLMs → Self-evolving LLM agents +1
Think Short, Defer Smart, Act, and Repeat: Calibrated Reasoning and Uncertainty-Aware Deferral for Edge LLM AgentsLLM reasoning → Reasoning for agents, planning and code +1
ReCo: Reweighting GRPO Against Distributional ConcentrationLLM reasoning → Reinforcement learning for reasoning
ToxScreen: Detecting Whether an LLM Has Been PoisonedLLM safety and security → Backdoor and data poisoning attacks
Thinking Under Uncertainty: Evidence Use and Information-Seeking in Language ModelsLLM reasoning → Reasoning benchmarks and evaluation
When Knowledge Changes: Metamorphic Testing of RAG Systems with MutationsRetrieval-augmented generation → RAG pipeline design and evaluation
Before Agents Speak: Pre-hoc Failure Risk Inference in Multi-Agent SystemsLLM agents and tool use → Medical and scientific domain agents
A First Look at Coding Agents' Compliance with AI Contribution Rules in Open-Source CommunitiesLLM agents and tool use → Software engineering coding agents
SecRespond: Benchmarking AI Agents for Real-World Post-Compromise Incident ResponseLLM agents and tool use → Agent threats, guardrails and governance
CheckVLA: Execution-Time Verification with Action-Conditioned World Model for Long-Horizon Mobile ManipulationWorld models → Embodied and robot world models +1
SkillRise: Agentic Reinforcement Learning for Cross-Task Skill EvolutionLLM agents and tool use → Agent skill learning and evolution
CodeSpec: Dual Executable Specifications for Agentic Long-Horizon Feature DevelopmentLLM agents and tool use → Software engineering coding agents
Do Latent Channels Actually Communicate? A Causal Audit of Latent Multi-Agent LLMLLM agents and tool use → General LLM agent frameworks and benchmarks
See2Think: Do Multimodal Models Really Use Intermediate Visual States?LLM reasoning → Multimodal reasoning +1
Metis: Memory Foundation ModelLLM reasoning → Reasoning for agents, planning and code
StatePlay: State-Aware Game World Models for Mechanics-Consistent GenerationWorld models → Video-generation world models and simulators

All 439