Papers

New on arXiv

Sun 30 Aug 2026

277 papers

Do World Models Make Better Robots? A Survey of Evaluation Benchmarks for Predictive Embodied IntelligenceWorld models → World models for navigation
Graph, Loop, and Harness Engineering for Zero-Trust Agentic Data Engineering and Analytical ProcessingLLM agents and tool use → Software engineering coding agents
To Think or Not to Think: Allocating Reasoning Where It HelpsLLM reasoning → Test-time compute and efficient thinking
Anatomy of a Decision: Uncertainty-aware Hierarchical Intent Learning via Flow Matching for Multimodal RecommendationDiffusion models → Diffusion recommenders
Dissecting Hierarchical Reasoning Models: A Mechanistic StudyLLM reasoning → Test-time compute and efficient thinking +1
AutoXRD: Autonomous LLM Agents and Comprehensive Evaluation for Powder Diffraction AnalysisLLM agents and tool use → Medical and scientific domain agents
Auditing Harness Tampering in Self-Improving AgentsSynthetic data and self-improvement for LLMs → Self-evolving LLM agents
Do Multimodal LLMs See Before They Read? Diagnosing Contextual SycophancyMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
Scientific Agent Skills: A Library of Procedural Knowledge for Research AgentsLLM agents and tool use → Web, search, and research agents
CUDA-Harness: Harnessing Agentic CUDA Kernel Generation and Optimization from Natural LanguageLLM agents and tool use → Software engineering coding agents
AgentProv: Auditing Agentic LLM API Providers via Tool-use Policy ProbesLLM agents and tool use → Agent threats, guardrails and governance
From Detection to Refusal: Safer LLMs via Circuit-Guided Weight ScalingLLM safety and security → Safety alignment, refusal and guardrails +1
Towards Agentic Cloud Engineering: Graph and Loop Engineering with a Zero-Trust Agent HarnessLLM agents and tool use → Software engineering coding agents
GUI-CC: Benchmarking Contextual Consistency of GUI World Models as Agent EnvironmentsWorld models → World model security and GUI agents +1
The Nearest Target Is the Wrong One: Target Separation in Arc2Face Identity UnlearningLLM safety and security → Machine unlearning
Tracing Generated Samples to Training-Data Clusters in Flow-Matching ModelsDiffusion models → Flow matching, time series and science
Budget-Aware Compression Pipeline for Single-GPU LLM Inference: Methods, Trade-offs, and Coupling EffectsEfficient LLM inference and serving → KV cache optimization
How do World Models and Policies Compose in LLM Agents? A Joint Spectral and Behavioral AccountWorld models → World models for language agents +1
Pak3H: Evaluating the Cost of Cultural Mismatch in LLM Alignment with a Human-Contextualized Urdu BenchmarkLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
Mitigating Over-Optimization in PRM-Guided Search in Mathematical Reasoning by Optimizing the GuideLLM reasoning → Reinforcement learning for reasoning +1
Can LLM Agents Discover? Evaluating Creativity on ML Engineering TasksLLM agents and tool use → Web, search, and research agents
Forget or Fine-tune? A Comparative Study of Machine Unlearning Strategies for Noisy Label CorrectionLLM safety and security → Machine unlearning
Reachability-Based Capability Confinement for LLM Agents under Indirect Prompt InjectionLLM agents and tool use → Agent threats, guardrails and governance +1
ActReal: System-Level Mobile Agents Challenge Mobile Automation DetectionLLM agents and tool use → GUI and mobile agents
The Intervention Gap in Latent World ModelsWorld models → Latent predictive world models
Discrete Diffusion Bridges for Spatiotemporally Aligned Image Translation and GenerationDiffusion models → Image generation, text-to-image and editing
Partition-Aware Unlearning for Removing Spurious Correlations in Large Vision-Language ModelsMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs +1
AutoCRAT: Within-trajectory Joint Control of Stochasticity and Compute for LLM ReasoningLLM reasoning → Test-time compute and efficient thinking
Evolutionary Soups: Evolving Mixture-of-Experts for Multi-Objective LLM AlignmentLLM alignment / RLHF / preference optimisation → LLM preference optimization methods +1
SpanCalib-VLM: Calibrated Hallucination Span Detection in Vision-Language ModelsMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
EDGE: Engine for Deterministic Graph Evaluation through Conversation Simulation from Graph Structured DSL ConfigurationLLM agents and tool use → General LLM agent frameworks and benchmarks
Training-Free Action Correction for VLA Model Failures via Language FeedbackVision-language-action models → Failure recovery and robustness
DataFoundry: Evolving Data Preparators via Recursive Self-ImprovementSynthetic data and self-improvement for LLMs → Code and instruction data synthesis
RIDGE: Region-Informed Derivative-Guided Evidence Selection for Long Video UnderstandingMultimodal LLMs / vision-language models → Video understanding with MLLMs
Detecting Hidden Chain-of-Thought in Large Language Models with Linguistic, Behavioral, and Mechanistic IndicatorsLLM reasoning → Prompting and chain-of-thought reasoning
SearchWiki: Learning to Build and Navigate Knowledge Wikis for Active Information SeekingRetrieval-augmented generation → Agentic RAG, agents and memory
On the Recoverability of Private Information Unlearning in Large Language ModelsLLM safety and security → Machine unlearning
Influence Is Not Authority: When Causal Guardrail Signals Make Legitimate Tool Use Look Like an Attack in Tool-Using LLM AgentsLLM agents and tool use → Agent threats, guardrails and governance +1
AcrossWAM1.0:A Modular Latent World-Action Stack for Compact Robot PoliciesWorld models → Embodied and robot world models
When Safety Speaks a Language: A Mechanistic Analysis of Safety-Language Identity Entanglement in LLMsLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +2
Compression-Aware Abstention: Teaching LLMs to Refuse When KV-Compression Masks Remove Answer EvidenceEfficient LLM inference and serving → KV cache optimization
Hallucination Mitigation for Large Vision-Language Models via Implicit Feature StabilizationMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
Towards Continual Test-Time Adaptation of Vision-Language Models in Open-Vocabulary Semantic SegmentationMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
Matrix-Game 3.5: Enhancing Real-Time Streaming Interactive World Models with Patch MemoryWorld models → Video-generation world models and simulators +1
Off-Manifold Refinement: Guiding Video Generators with a Frozen World ModelWorld models → Video-generation world models and simulators
VibeJam: A User Study Platform for Web Development with AgentsLLM agents and tool use → Software engineering coding agents
Improving Argument Saliency Coverage in Small LLMs for Long Legal Opinion Summarization via Sequence-Level DistillationKnowledge distillation → Knowledge distillation for LLMs
Perceive to Hypothesize, Verify to Ground: An Agentic Reasoning Framework for Open-World Geo-LocalizationLLM reasoning → Spatial and embodied reasoning +1
Uncertainty-Driven Replay Memory for Reinforcement LearningDeep reinforcement learning methods → Deep RL algorithms and representations
GenRubric: Self-Evolving Rubric Generation for Scalable LLM EvaluationSynthetic data and self-improvement for LLMs → LLM synthetic data generation
Influence-Directed Distillation: Solving the Diversity Bottleneck in Sampled-Token On-Policy DistillationKnowledge distillation → On-policy and self-distillation for LLMs
SkillForge: Compositional Skill Synthesis with Verification-in-the-Loop for Generating Formally Verified Dafny ProgramsLLM agents and tool use → Agent skill learning and evolution
EVAR: Evidence-Validated Hypothesis Admission for Budget-Aware Narrative ReasoningLLM reasoning → Reasoning benchmarks and evaluation
You Know What I Mean: A Benchmark for Agentic Conversational Reference GroundingLLM agents and tool use → General LLM agent frameworks and benchmarks
A^2Agent: Action-Aware Reinforcement Learning for Repository-Level Code Localization AgentsDeep reinforcement learning methods → RL training of language and agent models
SymVD: Symmetric Vision Language Action Distillation for Robot ManipulationVision-language-action models → Reinforcement learning and post-training +1
Structure Aware Neural Architecture Search for Mixture of ExpertsMixture-of-experts → MoE theory and gating functions
FRAMEWORKERS: A Dynamic Multi-Agent Framework for AI-Generated Video ProductionLLM agents and tool use → Multimodal and embodied agents
Foundation and Multimodal Large Language Models for Face Presentation and Morph Attack DetectionMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness
HiVe: Beyond Static Prompts for Multitask Learning via Hierarchy-based Vertical Mixture-of-ExpertsMixture-of-experts → Mixture of LoRA experts
InspectorGPT: A Comparative Reasoning Enhanced VLM for Comprehensive Industrial Anomaly DetectionLLM reasoning → Chain-of-thought reasoning for driving
Self-Aware Active Learning Enables Continual Improvement in Autonomous DrivingEnd-to-end autonomous driving → Reinforcement learning for driving
SmoothRL: Online Reinforcement Learning During Asynchronous ExecutionDeep reinforcement learning methods → RL training of language and agent models
GraM-Diff: A Unified Graph-Mamba Diffusion Framework for EEG-Based Alzheimer's Disease Data Generation and DiagnosisState-space sequence models (S4, Mamba) → Physiological signal modeling
PAGE-RAG: Provenance-Aware Graph Evidence Promotion for Fixed-Budget Multi-hop Retrieval-Augmented GenerationRetrieval-augmented generation → Knowledge-graph and structured-knowledge RAG
DriftingVLA: Native One-Step Vision-Language-Action Generation via Per-Dimension Temporal DriftingVision-language-action models → Efficient VLA inference and adaptation
ReTrace: Rejected-Trajectory Conditioning for Speculative DecodingEfficient LLM inference and serving → Speculative decoding
Cost-Effective Repository Exploration for Agentic Issue LocalizationLLM agents and tool use → Software engineering coding agents
Creation begins with understanding: LLMs as strategy designers for privacy-preserving tabular data synthesisSynthetic data and self-improvement for LLMs → Privacy-preserving tabular data synthesis
Evaluating a 4B open-weights local LLM for agentic DFT workflows: a literature reproducibility auditLLM agents and tool use → Medical and scientific domain agents
ACTD: Anchor-Based Cross-Tokenizer Distillation with Residual RegularizationLLM reasoning → Prompting and chain-of-thought reasoning +1
ButterMamba: Butterworth-Enhanced Spatial-Temporal Mamba for Efficient Traffic Flow PredictionState-space sequence models (S4, Mamba) → Video, motion and generative Mamba models
Detect Before You Attribute: Cascade Failure Attribution for Multi-Agent SystemsLLM agents and tool use → Software engineering coding agents
Harness-RL: Black-Box Reinforcement Learning with Action-Args Decoupling for Central-Agent Multi-Agent HarnessesLLM agents and tool use → Software engineering coding agents
InteractBench: Benchmarking LLMs on Competitive Programming under Unrevealed InformationLLM reasoning → Reasoning for agents, planning and code
SPLG-Mamba: Structure-Preserving Local-Global Mamba Network for Salient Object Detection in Optical Remote Sensing ImagesState-space sequence models (S4, Mamba) → Point clouds, detection and remote sensing
MI-Distillation: Selecting from Model-Interpolated Instruct-Reasoning Data Spectrum for Chain-of-Thought DistillationLLM reasoning → Prompting and chain-of-thought reasoning
AgenticRag-R1: Agentic Reinforcement Learning with Stack Memory for Multi-Step Reasoning, Retrieval and MemorizingLLM reasoning → Knowledge- and tool-augmented reasoning +2
CineForge: Self-Improving Agents for Long-Horizon Video GenerationLLM agents and tool use → Multimodal and embodied agents +1
JPO: Juris Policy Optimization for Structured Legal Reasoning in Criminal Judgment PredictionLLM reasoning → Reasoning for agents, planning and code
LLMs Interpret, Embeddings Organize, Graphs Emerge: Agent-Driven Compilation of Scientific KnowledgeLLM agents and tool use → Web, search, and research agents
Agent Zero Memory: Provenance-Aware Long-Term Memory for LLM AgentsLLM agents and tool use → Agent memory systems
Hindsight Memory-PRM: Supervising Memory Management with Auditable Hindsight CreditLLM agents and tool use → Agent memory systems
On the Resilience of Text-to-Video Diffusion Models to Hardware FaultsDiffusion models → Video generation
Towards a Systems Foundation for Agentic Skills: Architecture, Lifecycle, and SecurityLLM agents and tool use → General LLM agent frameworks and benchmarks
Guardrail-Agnostic Societal Bias Evaluation in Large Vision-Language ModelsMultimodal LLMs / vision-language models → Social bias and fairness in VLMs +1
Not Safe for All: Auditing the Dialect Penalty in Text-to-Image Safety PipelinesLLM safety and security → Safety alignment, refusal and guardrails
Call Neighbours Yourself: Graph Walks with Destination-Conditioned On-Policy Self-DistillationKnowledge distillation → On-policy and self-distillation for LLMs
Drive the Thoughts: Runtime Monitoring of VLA Reasoning-Trajectory ConsistencyLLM reasoning → Chain-of-thought reasoning for driving
SemTrace: Source-Grounded Semantic Signatures for Tracing LLM Exposure to Protected DocumentsLLM safety and security → Watermarking generated content and models
Which one is banana man? Evaluating vision-language models in multi-turn pragmatic interpretationMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
TACS: Trajectory-Aware Candidate Selection for LLM Jailbreak Suffix OptimizationLLM safety and security → Jailbreaking
PhysWave: Physics-Guided Latent Diffusion Models for Controllable Spatial Audio GenerationDiffusion models → Audio, speech and music
AGM: Achievement-Grounded Memory for Closed-Loop Agents with Frozen VLA PoliciesVision-language-action models → Reasoning and chain-of-thought in VLA +1
On the Plasticity Collapse in Continual Machine UnlearningLLM safety and security → Machine unlearning
Denoising as Projection: Constrained Optimization with Gradient-Guided DiffusionDiffusion models → Diffusion foundations and sampling
CoCoA: Context-Conditional Cultural Alignment for Large Language ModelsLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
GeoAgent: Evaluating VLM Geolocalization Through Embodied NavigationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
SIC-Agents: Benchmarking and Building an Adaptive Simulator for Pediatric Serious Illness Communication TrainingLLM agents and tool use → Medical and scientific domain agents
Not All Synthetic Data Are Equal: Expert-Committee Audit Screening for Imbalanced Crash-Injury-Severity Prediction in Automated Driving Systems

All 277