Papers

New on arXiv

Mon 21 Sep 2026

470 papers

Energy Time-Series Imputation with Differentially Private Diffusion Models via Clipping-Aware Objective ConditioningDiffusion models → Flow matching, time series and science
Less Uniform Discrete Diffusion is More Powerful and ScalableDiffusion models → Discrete and text diffusion
PanoFuse: Panorama-Enhanced Vision-Language-Action Learning with Decoupled Semantic-Geometric RoutingVision-language-action models → 3D and spatial-aware VLA
Agentic Video Understanding: A SurveyLLM agents and tool use → Multimodal and embodied agents
Silent Failures in Agent-Tool Interaction: An Audit of ToolUniverseLLM agents and tool use → Software engineering coding agents
Spec2COBOLRot: An Agentic-AI Degradation Loop for Realistic COBOL Corpus GenerationLLM agents and tool use → Software engineering coding agents
RGSQ: Riemannian Geometry-Sensitive Quantization for Large Vision-Language ModelsMultimodal LLMs / vision-language models → Visual token pruning and efficiency
RAG-NAROK: Retrieval-Aware Knowledge Corpus Poisoning in RAG with Source-specific RefutationRetrieval-augmented generation → Privacy and security of RAG
Rollout Efficiency in Reinforcement Learning for Reasoning Large Language Models: A Taxonomy and Future DirectionsLLM reasoning → Reinforcement learning for reasoning
Fast Recovery for LLM Serving via Decoupled Device Memory Lifetime in DynamoEfficient LLM inference and serving → Serving systems and request scheduling
Mean Velocity Matching: Rethinking Generative Dynamics in Diffusion ModelsDiffusion models → Flow matching, time series and science
Tipping Points in LLM-Based Multi-Agent Systems: Stance on Climate Change ActionLLM agents and tool use → Social simulation and multi-agent debate
Passes Alone, Fails Together: Benchmarking Semantic Coordination in Parallel LLM-Agent DevelopmentLLM agents and tool use → Software engineering coding agents
VLAQuantBench: Closed-Loop Evaluation of Post-Training Quantization for Vision-Language-Action ModelsVision-language-action models → VLA architectures and generalization
From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-ThoughtLLM reasoning → Prompting and chain-of-thought reasoning
TelecomGPT-R1: Unified Post-Training for Reasoning Across Heterogeneous Telecom TasksLLM reasoning → Reasoning for agents, planning and code
JAMB: Joint Action-Motion Diffusion for Bimanual ManipulationDiffusion models → Diffusion and flow policies for robotics
Uncertainty-Aware 3D Residual Wavelet Diffusion for Ultra Low-Field MRI Super-ResolutionDiffusion models → Medical imaging and segmentation
When LLM Agents Fail to Read the Room: ReAdapt for Relational Social ReasoningLLM reasoning → Reasoning for agents, planning and code +1
RULER: Instance-aware Rubric Rewards for SVG GenerationLLM alignment / RLHF / preference optimisation → Diffusion and multimodal model alignment
The AI Neuroscientist: An Interactive Agentic Interface for Neuroimaging AnalysisLLM agents and tool use → Medical and scientific domain agents
Trains but Doesn't Learn: A Post-Training Delivery Benchmark for LLM Agents as Forward-Deployed EngineersLLM agents and tool use → Software engineering coding agents
Indirect tipping: a social attack surface in AI agent populationsLLM agents and tool use → Agent threats, guardrails and governance
FinFIRST: Benchmarking Search Agents for Financial Information Retrieval, Sourcing and TraceabilityLLM agents and tool use → Financial trading and forecasting agents
X-Planner: Event-Structured Task Planning for Embodied IntelligenceVision-language-action models → Reasoning and chain-of-thought in VLA
Qwen-Audio-3.1-Realtime: Towards Reliable Agentic Voice InteractionLLM agents and tool use → General LLM agent frameworks and benchmarks
Attack Success Rate Is Not a Number: On Measurement Validity in Agentic AI Security EvaluationLLM agents and tool use → Agent threats, guardrails and governance
Mitigating Sequential Reappearance in Diffusion Data-Point UnlearningLLM safety and security → Vision-language and text-to-image safety
Who Pays for the KV Cache? Attributing Shared AI Inference Spend Across Kubernetes and LLM Provider BillsEfficient LLM inference and serving → Serving systems and request scheduling
WorldCrafter: Consistent Video World Model with Implicit 3D-aware MemoryWorld models → Video-generation world models and simulators
onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level CorrectionLLM agents and tool use → General LLM agent frameworks and benchmarks +1
DexTacWAM: A Visuo-Tactile World-Action Model for Dexterous ManipulationWorld models → Embodied and robot world models
Harness-Zero: Harness Distillation via Agent-as-HarnessLLM agents and tool use → Reinforcement learning for LLM agents
RRSI: Regularized Recursive Self-Improvement of Agent HarnessesLLM agents and tool use → Agent skill learning and evolution +1
Emergent Collusion in Long-Horizon LLM Agent InteractionLLM agents and tool use → Social simulation and multi-agent debate
Trajectory-Aware Benchmark Subset Selection for Cost-Efficient Software Engineering Agent Regression TestingLLM agents and tool use → Software engineering coding agents
PixelDiT2: Representation-Grounded Pixel Diffusion TransformersDiffusion models → Inverse problems and image restoration
SocioVerse2: A Longitudinal Dynamic Social Simulation Framework under a Human-AI Co-evolutionary ParadigmLLM agents and tool use → Social simulation and multi-agent debate
SLICEChat: Progressive In-Encoder Token Pruning for Whole-Slide Pathology Language ModelsMultimodal LLMs / vision-language models → Medical vision-language models +2
OSWorld-Pro: Process-based Evaluation for Computer Use AgentsLLM agents and tool use → GUI and mobile agents
SPHQuant: Efficient extreme low bit weight quantization for Vision-Language ModelsMultimodal LLMs / vision-language models → Visual token pruning and efficiency
DualWAM: Dual-System World Action Models for Asynchronous Global Planning and Local RefinementWorld models → Embodied and robot world models
When Tomorrow Becomes Today: Self-Evolving Policies for Agentic Time-Series ForecastingLLM agents and tool use → Financial trading and forecasting agents
Small-world Networks of Agents Brainstorm AI Risks to Support IdeationLLM agents and tool use → Social simulation and multi-agent debate
SPECTRA: Adaptive Execution of Speculative Decoding on a Runtime-Reconfigurable Tiled ArchitectureEfficient LLM inference and serving → Speculative decoding
PredActor: Predictive Action Diffusion for Steerable Onboard Humanoid ControlDiffusion models → Diffusion and flow policies for robotics
MedRSI: Recursive Self-Improvement for Medical Agents via Clinically Aligned Self-EvolutionLLM reasoning → Clinical and biomedical applications
GRUET: Quantifying Uncertainty of Agentic Reasoning-and-Acting ProcessesLLM reasoning → Reasoning for agents, planning and code +1
OPBackdoor: Opportunistic Backdoors via Alibi-Aligned ReasoningLLM safety and security → Backdoor and data poisoning attacks
INTCORT: Training-Free Spatial Reasoning Enhancement for Vision-Language Models via Input Transformations and Confidence RoutingLLM reasoning → Spatial and embodied reasoning +1
Decoding Guardrails: XAI-Guided Perturbation Analysis of Prompt Injection DetectionLLM safety and security → Prompt injection and agent security
Streaming Video Editing with Easy AdaptationDiffusion models → Video generation
PrismGPT: Proxy-Guided Learning for Region-Aware Photo Editing with Self-Synthesized ReasoningLLM reasoning → Multimodal reasoning
Construting Reverse Thinking: Developing Large Language Models' Reverse Thingking AbilityLLM reasoning → Prompting and chain-of-thought reasoning
D-JEPA: A Decision-Aligned Latent World ModelWorld models → Latent predictive world models
Beyond Visual Quality: A Study of Test-Time Planning with World Action ModelsWorld models → World models for language agents
Reasoning Topology Matters: A Controlled Study of LLM-Based Cybersecurity AnalysisLLM reasoning → Reasoning for agents, planning and code
Offline Reinforcement Learning for Distribution-Grid ProtectionOffline RL and imitation learning from data → Decision transformers and sequence modeling
Adapting Tree-Structured Speculative Decoding to DeepSeek-V4 for Efficient InferenceEfficient LLM inference and serving → Speculative decoding
What Makes a Good Medical Image Tokenizer? Rethinking Reconstruction and Generation in Medical Image TokenizationDiffusion models → Medical imaging and segmentation
Think Like a World Model, Act Like a VLA: Distilling World-Model Representations into Compact Robot PoliciesWorld models → Embodied and robot world models +1
DUMA-Bench: A Dual-Control Multi-Agent Benchmark for Evaluating LLM Agent SecurityLLM agents and tool use → Agent threats, guardrails and governance
Assessing Readability with LLMs: The Role of Reasoning and Few-Shot PromptingLLM reasoning → Reasoning benchmarks and evaluation
Analytical Power-Aware Provisioning for Prefill-Decode Disaggregated AI InferenceEfficient LLM inference and serving → Serving systems and request scheduling
Written as a Record, Read as an Address: What a Forward Pass Leaves in an Operation's KV CacheEfficient LLM inference and serving → KV cache optimization
From Semantic Decisions to Feasible Trajectories: Self-Evolving LLM-Guided Optimal Control for Narrow-Space ParkingSynthetic data and self-improvement for LLMs → Self-evolving LLM agents
Relationally Grounded Latent World Models for Autonomous DrivingWorld models → World models for autonomous driving +1
Ascent: An Agentic System over the Model Context Protocol for Real-World Clinical Data AnalysisLLM agents and tool use → Medical and scientific domain agents
What do VLM-Based Vision-Language Navigation Models Rely on: Interpreting and Steering Policy BehaviorMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
HyperCLIP++: Fine-tuning CLIP forOpen-vocabulary Semantic Segmentation in Hyperbolic SpaceMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
ARSTAG: An Agentic Real2Sim2Real System for Task-Specific Robot Data GenerationLLM agents and tool use → Multimodal and embodied agents
ME-VLM: A Unified VLM for Embodied Cognition and Agent CoordinationLLM agents and tool use → Multimodal and embodied agents
Bridge3D: Enabling Vision-Language-Action Models to See and Act in 3DVision-language-action models → 3D and spatial-aware VLA
CMAMBADEPTH: Self-supervised Monocular Depth Estimation with Channel Mamba and Hybrid AttentionState-space sequence models (S4, Mamba) → Vision Mamba backbones
Lifted Bellman Linear Programming for Offline Reinforcement LearningOffline RL and imitation learning from data → Offline RL algorithms
VPRune: Efficient Training-free Pre-LLM Visual Token PruningMultimodal LLMs / vision-language models → Visual token pruning and efficiency +1
Fathom-Vaidya: Advancing Medical Reasoning with Rubric-Based RewardsLLM reasoning → Clinical and biomedical applications
Spatial Action Review: A Visual Analytics Dashboard for Auditing Language-to-Action Hand-offs in Electron MicroscopyMultimodal LLMs / vision-language models → Chart understanding and anomaly detection
RAILS: Retrieval-Augmented Incremental LLM Clustering at ScaleRetrieval-augmented generation → Generation and context integration
ActGov: Governing LLM Agent Actions via Policy-Constrained ValidationLLM agents and tool use → Agent threats, guardrails and governance
Comparing Latent Concept Formation in State Space Models and Transformers via Sparse AutoencodersState-space sequence models (S4, Mamba) → Core SSM architectures for language
FoldQuantVLA: Native Low-Bit Quantization of Vision-Language-Action Models via Consistent FoldingVision-language-action models → Efficient VLA inference and adaptation
Information-Time Proximal Policy OptimizationLLM reasoning → Reinforcement learning for reasoning
On the Information-Theoretic Limits of Latent-Space Watermarking Through Pretrained GeneratorsLLM safety and security → Watermarking generated content and models
URA-NER: A Unified Retrieval-Augmented Framework with Retrieval Alignment and Uncertainty Reduction for Low-Resource NERRetrieval-augmented generation → Generation and context integration
VLM-in-Sandbox: Visual Workspaces for Agentic Visual ReasoningLLM reasoning → Multimodal reasoning +2
Dissecting Agentic Forensics: The Role of Triage, Prompting, and Evidence Arbitration in Open-World Fake Image DetectionLLM agents and tool use → Agent threats, guardrails and governance
A Lean and Spec-Driven AI-Assisted Software Development Lifecycle for Applied AI Education: The AI-SDLC ApproachLLM agents and tool use → Software engineering coding agents
LADDER: Graph-Guided Diffusion Language Models for Efficient Multi-Hop ReasoningLLM reasoning → Knowledge- and tool-augmented reasoning +3
LiAuto-MindViT: A Hybrid Vision Backbone with Adaptive Bidirectional MambaState-space sequence models (S4, Mamba) → Vision Mamba backbones
NeuIDO: Neural Intrinsic Dynamics Operator for Physics-Informed 4D World ModelsWorld models → Latent predictive world models
HappyWorld-BenchWorld models → Video-generation world models and simulators
KV-COBRA: KV Cache Compression via Co-Optimized Bit-Rank AllocationEfficient LLM inference and serving → KV cache optimization
When and How Should an Agent Clarify? CIGAsk: Teaching LLMs to Clarify via Counterfactual Information GainLLM agents and tool use → Web, search, and research agents
Classifier-Free Guidance in Flow Matching: Non-Autonomous Potentials, Overshoot, and Posterior-Mean ControlDiffusion models → Flow matching, time series and science
How Many Pixels Is a Digit Worth? Place-Aware Coordinate Entropy for GUI Agent Confidence EstimationLLM agents and tool use → GUI and mobile agents
Hierarchical Prompt Learning for Hyperbolic Vision-Language ModelsMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
Unsupervised Brain Anomaly Detection as a Bayesian Inverse Problem with Diffusion PriorDiffusion models → Medical imaging and segmentation
Canonical Procedural Actions: An Auditable Annotation Protocol for Tool-Use Agent TracesLLM agents and tool use → Software engineering coding agents
MemCalib: Benchmarking and Optimizing Memory Use in LLM AgentsLLM agents and tool use → Agent memory systems

All 470