Papers

New on arXiv

Sun 10 May 2026

352 papers

Skill-Contracted Agents for Evidence-Aware Materials Literature AnalysisRetrieval-augmented generation → Agentic RAG, agents and memory
Shapley Context Pruning: A Cooperative Game Perspective for Context Reranking and PruningRetrieval-augmented generation → Generation and context integration
PEEK: Predictive Queue-Informed KV Cache Management for LLM ServingEfficient LLM inference and serving → KV cache optimization
MiroBench: Benchmarking Realism in Agentic Simulation of Real-world DiscussionsLLM agents and tool use → Social simulation and multi-agent debate
Lost or Hidden? A Concept-Level Forgetting in Supervised Continual LearningMechanistic interpretability → SAE features and their applications
GeoSym127K: Scalable Symbolically-verifiable Synthesis for Multimodal Geometric ReasoningLLM reasoning → Spatial and embodied reasoning +1
Fre-Res: Frequency-Residual Video Token Compression for Efficient Video MLLMsEfficient LLM inference and serving → Visual token pruning in multimodal models
Portable Agent Memory: A Protocol for Cryptographically-Verified Memory Transfer Across Heterogeneous AI AgentsLLM agents and tool use → Agent memory systems
From Code-Centric to Intent-Centric Software Engineering: A Reflexive Thematic Analysis of Generative AI, Agentic Systems, and Engineering AccountabilityLLM agents and tool use → Software engineering coding agents
AgentShield: Deception-based Compromise Detection for Tool-using LLM AgentsLLM agents and tool use → Agent threats, guardrails and governance +1
A Switching System Theory of Q-Learning with Linear Function ApproximationDeep reinforcement learning methods → RL theory and sample complexity
Trust Region Inverse Reinforcement Learning: Explicit Dual Ascent using Local Policy UpdatesOffline RL and imitation learning from data → Inverse reinforcement learning +1
Efficient LLM Reasoning via Variational Posterior Guidance with Efficiency AwarenessLLM reasoning → Reinforcement learning for reasoning
LoopUS: Recasting Pretrained LLMs into Looped Latent Refinement ModelsLLM reasoning → Test-time compute and efficient thinking
ACSAC: Adaptive Chunk Size Actor-Critic with Causal Transformer Q-NetworkOffline RL and imitation learning from data → Robot manipulation imitation learning +1
An Execution-Verified Multi-Language Benchmark for Code Semantic ReasoningLLM reasoning → Reasoning for agents, planning and code
DisagMoE: Computation-Communication overlapped MoE Training via Disaggregated AF-Pipe ParallelismMixture-of-experts → Efficient MoE training and inference systems
MT-JailBench: A Modular Benchmark for Understanding Multi-Turn Jailbreak AttacksLLM safety and security → Jailbreaking
CalBench: Evaluating Coordination-Privacy Trade-offs in Multi-Agent LLMsLLM agents and tool use → Agent threats, guardrails and governance
Oracle Poisoning: Corrupting Knowledge Graphs to Weaponise AI Agent ReasoningLLM agents and tool use → Agent threats, guardrails and governance
Dystruct: Dynamically Structured Diffusion Language Model Decoding via Bayesian InferenceDiffusion models → Discrete and text diffusion +1
Learning to Compress Time-to-Control: A Reinforcement Learning Framework for Chronic Disease ManagementOffline RL and imitation learning from data → Safe offline RL +1
Evaluating Tool Cloning in Agentic-AI EcosystemsLLM agents and tool use → Software engineering coding agents
LEAD: Length-Efficient Adaptive and Dynamic Reasoning for Large Language ModelsLLM reasoning → Test-time compute and efficient thinking
CrossVL: Complexity-Aware Feature Routing and Paired Curriculum for Cross-View Vision-Language DetectionMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
LLM Agents Enable User-Governed Personalization Beyond Platform BoundariesLLM agents and tool use → Agent memory systems
EvoPref: Multi-Objective Evolutionary Optimization Discovers Diverse LLM Alignments Beyond Gradient DescentLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Exploitation Without Deception: Dark Triad Feature Steering Reveals Separable Antisocial Circuits in Language ModelsMechanistic interpretability → Activation steering methods
Marrying Generative Model of Healthcare Events with Digital Twin of Social Determinants of Health for Disease ReasoningDiffusion models → Medical imaging and segmentation
Primal-Dual Guided Decoding for Constrained Discrete DiffusionDiffusion models → Discrete and text diffusion
TIDES: Implicit Time-Awareness in Selective State Space ModelsState-space sequence models (S4, Mamba) → Core SSM architectures for language
KV-RM: Regularizing KV-Cache Movement for Static-Graph LLM ServingEfficient LLM inference and serving → KV cache optimization
One for All: A Non-Linear Transformer can Enable Cross-Domain Generalization for In-Context Reinforcement LearningDeep reinforcement learning methods → Deep RL algorithms and representations
On-Policy Distillation with Best-of-N Teacher Rollout SelectionKnowledge distillation → On-policy and self-distillation for LLMs +1
Distilling 3D Spatial Reasoning into a Lightweight Vision-Language Model with CoTLLM reasoning → Spatial and embodied reasoning +2
MOTOR-Bench: A Real-world Dataset and Multi-agent Framework for Zero-shot Human Mental State UnderstandingLLM agents and tool use → Multimodal and embodied agents
Calibrate, Don't Curate: Label-Efficient Estimation from Noisy LLM JudgesLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
DriveFuture: Future-Aware Latent World Models for Autonomous DrivingWorld models → World models for autonomous driving +1
Ambig-DS: A Benchmark for Task-Framing Ambiguity in Data-Science AgentsLLM agents and tool use → Tool learning and function calling
Discriminative Span as a Predictor of Synthetic Data Utility via Classifier ReconstructionSynthetic data and self-improvement for LLMs → Model collapse from synthetic data
ConFixGS: Learning to Fix Feedforward 3D Gaussian Splatting with Confidence-Aware Diffusion Priors in Driving ScenesDiffusion models → 3D, point cloud and scene generation
MonitoringBench: Semi-Automated Red-Teaming for Agent MonitoringLLM agents and tool use → Agent threats, guardrails and governance +1
Forcing-KV: Hybrid KV Cache Compression for Efficient Autoregressive Video Diffusion ModelsDiffusion models → Video generation +1
DeepTumorVQA: A Hierarchical 3D CT Benchmark for Stage-Wise Evaluation of Medical VLMs and Tool-Augmented AgentsMultimodal LLMs / vision-language models → Medical vision-language models
Absurd World: A Simple Yet Powerful Method to Absurdify the Real-world for Probing LLM Reasoning CapabilitiesLLM reasoning → Reasoning benchmarks and evaluation
CodeClinic: Evaluating Automation of Coding Skills for Clinical Reasoning AgentsLLM reasoning → Clinical and biomedical applications +1
MedMeta: A Benchmark for LLMs in Synthesizing Meta-Analysis Conclusion from Medical StudiesRetrieval-augmented generation → Medical and biomedical RAG
Metropolis-Adjusted Diffusion ModelsDiffusion models → Diffusion foundations and sampling
Workspace Optimization: How to Train Your AgentLLM agents and tool use → Agent skill learning and evolution
Make Each Token Count: Towards Improving Long-Context Performance with KV Cache EvictionEfficient LLM inference and serving → KV cache optimization
"Training robust watermarking model may hurt authentication!'' Exploring and Mitigating the Identity Leakage in Robust WatermarkingLLM safety and security → Watermarking generated content and models
Plan2Cleanse: Test-Time Backdoor Defense via Monte-Carlo Planning in Deep Reinforcement LearningDeep reinforcement learning methods → Adversarial robustness and RL in finance
Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language ModelsEfficient LLM inference and serving → KV cache optimization
Any2Any 3D Diffusion Models with Knowledge Transfer: A Radiotherapy Planning StudyDiffusion models → Medical imaging and segmentation
Statistical Scouting Finds Debate-Safe but Not Debate-Useful Cases: A Matched-Ceiling Study of Open-Weight LLM Reasoning ProtocolsLLM reasoning → Reasoning for agents, planning and code +1
Reflection Anchors for Propagation-Aware Visual Retention in Long-Chain Multimodal ReasoningLLM reasoning → Multimodal reasoning
Byte-Exact Deduplication in Retrieval-Augmented Generation: A Three-Regime Empirical Analysis Across Public BenchmarksRetrieval-augmented generation → Generation and context integration
Edit-Based Refinement for Parallel Masked Diffusion Language ModelsDiffusion models → Discrete and text diffusion
SoccerLens: Grounded Soccer Video Understanding Beyond AccuracyMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Neuromorphic Reinforcement Learning for Quadruped Locomotion Control on Uneven TerrainDeep reinforcement learning methods → RL for robot locomotion and manipulation
Trust Me, Import This: Dependency Steering Attacks via Malicious Agent SkillsLLM agents and tool use → Agent threats, guardrails and governance
DeformMaster: An Interactive Physics-Neural World Model for Deformable Objects from VideosWorld models → Latent predictive world models
CLR-voyance: Reinforcing Open-Ended Reasoning for Inpatient Clinical Decision Support with Outcome-Aware RubricsLLM reasoning → Clinical and biomedical applications
ConCovUp: Effective Agent-Based Test Driver Generation for Concurrency TestingLLM agents and tool use → Software engineering coding agents
When Adaptation Fails: A Gradient-Based Diagnosis of Collapsed Gating in Vision-Language Prompt LearningMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
Crosslingual On-Policy Self-Distillation for Multilingual ReasoningLLM reasoning → Prompting and chain-of-thought reasoning +1
TIDE-Bench: Task-Aware and Diagnostic Evaluation of Tool-Integrated ReasoningLLM reasoning → Reasoning for agents, planning and code +1
TacoMAS: Test-Time Co-Evolution of Topology and Capability in LLM-based Multi-Agent SystemsLLM agents and tool use → General LLM agent frameworks and benchmarks
Drift is a Sampling Error: SNR-Aware Power Distributions for Long-Horizon Robotic PlanningVision-language-action models → Reinforcement learning and post-training
Governing AI-Assisted Security Operations: A Design Science Framework for Operational Decision SupportLLM agents and tool use → Agent threats, guardrails and governance
MemPrivacy: Privacy-Preserving Personalized Memory Management for Edge-Cloud AgentsLLM agents and tool use → Agent threats, guardrails and governance
Mixture of Layers with Hybrid AttentionMixture-of-experts → Expert routing strategies
Uncertainty-Aware and Decoder-Aligned Learning for Video SummarizationLLM alignment / RLHF / preference optimisation → Diffusion and multimodal model alignment
EpiGraph: Building Generalists for Evidence-Intensive Epilepsy Reasoning in the WildLLM reasoning → Clinical and biomedical applications
AI Security Policy Should Assess Systems, Not Only ModelsLLM agents and tool use → Agent threats, guardrails and governance +1
PermuQuant: Lowering Per-Group Quantization Error by Reordering Channels for Diffusion ModelsDiffusion models → Diffusion foundations and sampling
Hidden Error Awareness in Chain-of-Thought Reasoning: The Signal Is Diagnostic, Not CausalLLM reasoning → Prompting and chain-of-thought reasoning +1
Don't Click That: Teaching Web Agents to Resist Deceptive InterfacesLLM agents and tool use → Agent threats, guardrails and governance
Beyond Language: Format-Agnostic Reasoning Subspaces in Large Language ModelsLLM reasoning → Reasoning benchmarks and evaluation
Not All Thoughts Need HBM: Semantics-Aware Memory Hierarchy for LLM ReasoningLLM reasoning → Test-time compute and efficient thinking +1
Outlier-Robust Diffusion Solvers for Inverse ProblemsDiffusion models → Inverse problems and image restoration
Beyond Position Bias: Shifting Context Compression from Position-Driven to Semantic-DrivenEfficient LLM inference and serving → Prompt and context compression
SpaceMind++: Toward Allocentric Cognitive Maps for Spatially Grounded Video MLLMsMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Through the Lens of Character: Resolving Modality-Role Interference in Multimodal Role-Playing AgentMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning +1
Inverse Design for Conditional Distribution MatchingDiffusion models → Diffusion foundations and sampling
Offline Preference Optimization for Rectified Flow with Noise-Tracked PairsDiffusion models → Diffusion and flow policies for robotics +1
Evading Visual Aphasia: Contrastive Adaptive Semantic Token Pruning for Vision-Language ModelsMultimodal LLMs / vision-language models → Visual token pruning and efficiency +1
AtteConDA: Attention-Based Conflict Suppression in Multi-Condition Diffusion Models and Synthetic Data AugmentationDiffusion models → Image generation, text-to-image and editing
SimWorld Studio: Automatic Environment Generation with Evolving Coding Agent for Embodied Agent LearningLLM agents and tool use → Multimodal and embodied agents
Perception Without Engagement: Dissecting the Causal Discovery Deficit in LMMsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
MACAA: Belief-Revision Multi-Agent Reasoning for Code Authorship VerificationLLM reasoning → Reasoning benchmarks and evaluation +1
From Passive Reuse to Active Reasoning: Grounding Large Language Models for Neuro-Symbolic Experience ReplayLLM reasoning → Reinforcement learning for reasoning
RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action ModelsVision-language-action models → Failure recovery and robustness
Sparsity Moves Computation: How FFN Architecture Reshapes Attention in Small TransformersMixture-of-experts → Expert routing strategies
BadDLM: Backdooring Diffusion Language Models with Diverse TargetsDiffusion models → Robustness, privacy and adversarial aspects +1
Empowering VLMs for Few-Shot Multimodal Time Series Classification via Tailored Agentic ReasoningLLM agents and tool use → Financial trading and forecasting agents
Do Linear Probes Generalize Better in Persona Coordinates?Mechanistic interpretability → Activation steering methods
Kinetic-Optimal Scheduling with Moment Correction for Metric-Induced Discrete Flow Matching in Zero-Shot Text-to-SpeechDiffusion models → Audio, speech and music
LiteMedCoT-VL: Parameter-Efficient Adaptation for Medical Visual Question AnsweringMultimodal LLMs / vision-language models → Medical vision-language models +1
31.1 A 14.08-to-135.69Token/s ReRAM-on-Logic Stacked Outlier-Free Large-Language-Model Accelerator with Block-Clustered Weight-Compression and Adaptive Parallel-Speculative-DecodingEfficient LLM inference and serving → Speculative decoding

All 352