Papers

New on arXiv

Fri 26 Jun 2026

359 papers

Position Is All You Need: A Free Lunch Token Compression Strategy for MLLM-based Referring Expression SegmentationEfficient LLM inference and serving → Visual token pruning in multimodal models
AdaThinking-E: One-Token Entropy Regulation for Adaptive ThinkingLLM reasoning → Test-time compute and efficient thinking +1
AI Learning and Conceptual Transfer in the Game of Hidden RulesDeep reinforcement learning methods → RL for games and combinatorial optimization
GLIDE: Guided Layerwise Hybrid Attention for Efficient LLM InferenceEfficient LLM inference and serving → KV cache optimization
ARdena: Scenario-driven control of real-time LLM agentsLLM agents and tool use → General LLM agent frameworks and benchmarks
LeanFlow: A Case Study in Workflow-Driven Lean AutoformalizationLLM agents and tool use → Educational agents and math problem solving
Self-Evolving Just-In-Time Memory for Proactive Embodied SafetyMultimodal LLMs / vision-language models → Embodied and 3D vision-language models +2
CIGPO: Contextual Information-Gain Policy Optimization for Multi-Turn Evidence-Reading LLM AgentsLLM reasoning → Reinforcement learning for reasoning +1
TRACE: Trajectory-Based Safety Patch Learning for LLM Post-Training RealignmentLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
Normalized Rewards for Preference OptimizationLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
DELTAVID: Enhancing Fine-Grained Spatiotemporal Perception with Cross-Video DifferencesMultimodal LLMs / vision-language models → Video understanding with MLLMs
RIPA: Sensory-Vector Prompt Injection Attacks on LLM-Controlled ROS 2 RobotsLLM safety and security → Prompt injection and agent security
Physics-Grounded Disentangled Flow Modeling for Brain Disease Progression TrajectoryDiffusion models → Medical imaging and segmentation
What LLMs explain is not what they believe: Evaluating explanation sufficiency under models' own input beliefsLLM reasoning → Reasoning benchmarks and evaluation
Animation2Code: Evaluating Temporal Visual Reasoning in Video-to-Code GenerationLLM reasoning → Multimodal reasoning +1
Search for Truth from Reasoning: A Dynamic Representation Editing Framework for Steering LLM TrajectoriesLLM reasoning → Test-time compute and efficient thinking
Correct codes for the wrong reasons? validating LLMs as measurement instruments for theoretical constructsLLM reasoning → Reasoning benchmarks and evaluation
R2-Searcher: Calibrating Retrieval and Reasoning Boundaries for Agentic SearchLLM reasoning → Knowledge- and tool-augmented reasoning +1
SEAD: Competence-Aware On-Policy Distillation via Entropy-Guided SupervisionKnowledge distillation → On-policy and self-distillation for LLMs
IMCBench: A benchmark for multimodal LLMs in Image-grounded Medical ConversationsLLM reasoning → Clinical and biomedical applications +1
DataComp-VLM: Improved Open Datasets for Vision-Language ModelsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Turn-Averaged SAEs for Feature Discovery and Long-Context AttributionMechanistic interpretability → SAE features and their applications
MammoFlow: Multiview Mammogram Synthesis with Anatomically Consistent Flow MatchingDiffusion models → Medical imaging and segmentation
A Gravitational Interpretation of Safety Reversion under Fine-TuningLLM safety and security → Safety alignment, refusal and guardrails
Detecting Clinical Hallucinations in LVLMs via Counterfactual Visual Grounding UncertaintyMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
CLEAR-MoE: Shared-Basis Expert Extraction from Frozen Vision Transformers via Calibration-Driven Layer SelectionMixture-of-experts → Vision mixture-of-experts
HDDPM: Heteroscedastic Denoising Diffusion Probabilistic Model for Quantitative Low-Count Brain PET RecoveryDiffusion models → Medical imaging and segmentation
TUA-Bench: A Benchmark for General-Purpose Terminal-Use AgentsLLM agents and tool use → GUI and mobile agents
An Agentic AI Pipeline for Appliance-Level Energy Anomaly Detection and LLM-Driven RecommendationsLLM agents and tool use → Financial trading and forecasting agents
Is Lying an Emergent Behaviour in LLMs? Evidence from Gaslighting AI agents in a Sustainability GameLLM agents and tool use → Social simulation and multi-agent debate
Event-Conditioned Diagnostics of Kinematic, Contact, and Object-Permanence Structure in Passive Object-State World ModelsWorld models → Latent predictive world models
LLM agents security duality: a comprehensive survey of self-security and empowered cybersecurityLLM agents and tool use → Agent threats, guardrails and governance
SemFlowRAG: Directed Semantic Flow from Abstraction to Evidence for Complex ReasoningLLM reasoning → Knowledge- and tool-augmented reasoning +1
Domain-Informed Multi-View Self-Distillation for Astronomical Light-Curve Representation Learning with JEPAKnowledge distillation → General knowledge distillation methods
When AI Reviews Its Own Code: Recursive Self-Training Collapse in Code LLMsSynthetic data and self-improvement for LLMs → Code and instruction data synthesis
Dockerless: Environment-Free Program Verifier for Coding AgentsLLM agents and tool use → Software engineering coding agents +1
SWE-MeM: Learning Adaptive Memory Management for Long-Horizon Coding AgentsLLM agents and tool use → Software engineering coding agents
Position: RL Researchers Need to Distinguish Between Solving Simulators and Using Simulators as a ProxyDeep reinforcement learning methods → RL for robot locomotion and manipulation
Building to the Test: Coding Agents Deliver What You Check, Not What You RequestedLLM agents and tool use → Software engineering coding agents
WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data CurationOffline RL and imitation learning from data → Robot manipulation imitation learning
VGB for Masked Diffusion Model: Efficient Test-time Scaling for Reward Satisfaction and Sample EditingLLM reasoning → Diffusion reasoning and test-time scaling +1
Democratic ICAI: Debating Our Way to Steering Principles from PreferencesLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
Agentic Hardware Design as Repository-Level Code EvolutionLLM agents and tool use → Software engineering coding agents
Vision-Default, Prior-Override: Causal Mechanisms of Perception-Knowledge Conflict in Vision-Language ModelsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
RSICCLLM: A Multimodal Large Language Model for Remote Sensing Image Change CaptioningMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
HPRO: Hierarchical Progressive Reward Optimization via Preference Extraction for Emotional Text-to-SpeechLLM alignment / RLHF / preference optimisation → Preference alignment for recommenders and RAG
Govern the Repository, Not the Agent: Measuring Ecosystem-Level Risk in AI-Native SoftwareLLM agents and tool use → Software engineering coding agents
Exposure Bias Can Alleviate Itself via Directional and Frequency Rectification in Flow MatchingDiffusion models → Flow matching, time series and science
GBC: Gradient-Based Connections for Optimizing Multi-Agent SystemsLLM agents and tool use → General LLM agent frameworks and benchmarks
Cognitive Episodes in LLM Reasoning Traces Enable Interpretable Human Item Difficulty PredictionLLM reasoning → Test-time compute and efficient thinking
LLawCo: Learning Laws of Cooperation for Modeling Embodied Multi-Agent BehaviorLLM agents and tool use → General LLM agent frameworks and benchmarks
Tandem Reinforcement Learning with Verifiable RewardsLLM reasoning → Reinforcement learning for reasoning
EchoSonar-R: A Multi-View Reasoning-Enabled Model for Disease Classification and Report Generation in EchocardiographyLLM reasoning → Clinical and biomedical applications
Robust Harmful Features Under Jailbreak Attacks: Mechanistic Evidence from Attention Head Specialization in Large Language ModelsLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
Regularized Reward-Punishment Reinforcement LearningDeep reinforcement learning methods → Deep RL algorithms and representations
Monocular Avatar Reconstruction via Cascaded Diffusion Priors and UV-Space Differentiable ShadingDiffusion models → 3D, point cloud and scene generation
Beyond Sparse Supervision: Diffusion-Guided Learning for Few-Shot Graph Fraud DetectionDiffusion models → Flow matching, time series and science
From Tokens to States: LLMs as a Special Case of World Models and the Continuous Path BeyondWorld models → World models for language agents
Diffusion Model Attribution via Spectral Coupling of Denoiser ResponsesDiffusion models → Diffusion foundations and sampling
ToolPrivacyBench: Benchmarking Purpose-Bound Privacy in Tool-Using LLM AgentsLLM agents and tool use → Agent threats, guardrails and governance
Can LLMs Judge Better Than They Generate? Evaluating Task Asymmetry, Mechanistic Interpretability and Transferability for In-Context QAMechanistic interpretability → Probing deception, bias and beliefs
AirGroundBench: Probing Spatial Intelligence in Multimodal Large Models under Heterogeneous Multi-View Embodied CollaborationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
TempAct: Advancing Temporal Plausibility in Autoregressive Video Generation via Planner-Executor RLDiffusion models → Video generation
From Detection to Action: Using LLM Agents for Fault-Tolerant ControlLLM agents and tool use → General LLM agent frameworks and benchmarks
HumanMoveVQA: Can Video MLLMs reason about human movement in videos?LLM reasoning → Spatial and embodied reasoning +1
AdvancedShelLM: A Stateful Multi-Agent LLM Honeypot for SSH DeceptionLLM agents and tool use → Agent threats, guardrails and governance
Latent Visual Diffusion Reasoning with Monte Carlo Tree SearchLLM reasoning → Multimodal reasoning
ToxiREX: A Dataset on Toxic REasoning in ConteXtLLM reasoning → Reasoning benchmarks and evaluation
Directing the World: Fast Autoregressive Video Generation with Compositional Human-Camera ControlWorld models → Video-generation world models and simulators
Understanding How MLLMs Describe Artworks Using Token Activation MapsLLM reasoning → Multimodal reasoning +1
VASAE: Naming SAE Dictionary Directions with Vocabulary-Aligned AnchoringMechanistic interpretability → SAE features and their applications
Agentic AI-Powered Re-Identification: An Emerging, Scalable Threat to Mobility Microdata PrivacyLLM agents and tool use → Agent threats, guardrails and governance
Controllable Histopathology Image Synthesis with Training-free Structural Initialization and Textural ModulationDiffusion models → Medical imaging and segmentation
When Multi-Robot Systems Meet Agentic AI:Towards Embodied Collective IntelligenceLLM agents and tool use → Multimodal and embodied agents
S2-VLA: State-Space Guided Vision-Language-Action Models for Long-Horizon ManipulationVision-language-action models → 3D and spatial-aware VLA
LocalNav: Distilling Frontier VLMs and Embodied RL for On-Device Object Goal NavigationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
FlexMoE: One-for-All Nested Intra-Expert Pruning for MoE Language ModelsMixture-of-experts → MoE large language models
PPO-EAL: Exact Augmented Lagrangian Proximal Policy Optimization for Safe Robotic ControlDeep reinforcement learning methods → Safe and constrained RL
CSD: Content-aware Speculative Decoding for Efficient Image GenerationEfficient LLM inference and serving → Fast autoregressive visual generation
Video-MME-Logical: A Controlled Diagnostic Benchmark for Video Temporal-Logical ReasoningLLM reasoning → Multimodal reasoning +1
NormAct: Benchmarking Embodied Agents' Proactive Compliance with Unspoken Social NormsMultimodal LLMs / vision-language models → Embodied and 3D vision-language models +1
ATOD: Annealed Turn-Aware On-Policy Distillation for Multi-Turn Agentic TasksLLM agents and tool use → Reinforcement learning for LLM agents +1
SpikeVLA: Vision-Language-Action Models with Spiking Neural NetworksVision-language-action models → Generalist VLA models and benchmarks
Agent vs. Parametric World Models: Hybrid Planning for Reliable Language AgentsWorld models → World models for language agents +1
Optimizing Teacher-Student Partitioning for Scalable Knowledge Distillation on HPC SystemsKnowledge distillation → General knowledge distillation methods
SHIFT: Gate-Modulated Activation Steering for Knowledge Conflict Mitigation in Retrieval-Augmented GenerationRetrieval-augmented generation → RAG robustness and adaptive retrieval +1
Understanding Rollout Error in Graph World ModelsWorld models → World models for language agents
ModaFlow: Modality-Aware Flow Matching for High-Fidelity Virtual Try-OnDiffusion models → 3D, point cloud and scene generation
NormGuard: Reward-Preserving Norm Constraints in Flow-Matching Reinforcement LearningDiffusion models → Diffusion and flow policies for robotics +2
RS-Diffuser: Risk-Sensitive Diffusion Planning with Distributional Value GuidanceOffline RL and imitation learning from data → Diffusion and flow policies for offline RL +1
PixelU: A U-Shaped Transformer for Efficient End-to-End Pixel DiffusionDiffusion models → Diffusion foundations and sampling
Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?Vision-language-action models → Efficient VLA inference and adaptation
SIFT: Self-Imagination Fine-Tuning for Physically Plausible Motion in Video Diffusion ModelsDiffusion models → Video generation
The Weakest Link Tells It All: Outcome-Supervised Process Reward Modeling via Learnable Credit AssignmentLLM reasoning → Reinforcement learning for reasoning +1
ToE: A Hierarchical and Explainable Claim Verification Framework with Dynamic Multi-source Evidence Retrieval and AggregationLLM reasoning → Reasoning benchmarks and evaluation
BashCoder-R1: Towards Robust and Explainable Bash Code Generation with Robustness-Aware Group Relative Policy OptimizationLLM reasoning → Reasoning for agents, planning and code
Bifocal Diffusion Language Models: Asymmetric Bidirectional Context for Parallel GenerationDiffusion models → Discrete and text diffusion +2
Learning to Reason with Curriculum II: Compositional GeneralizationLLM reasoning → Mathematical reasoning
Low-Agreeableness Persona Conditioning for Safe LLM Fine-TuningLLM safety and security → Safety alignment, refusal and guardrails
Class-frequency Guided Noise Schedule for Diffusion ModelsDiffusion models → Diffusion foundations and sampling

All 359