Papers

New on arXiv

Mon 6 Apr 2026

391 papers

Targeted Remasking: Replacing Token Editing with Token-to-Mask Refinement in Discrete Diffusion Language ModelsDiffusion models → Discrete and text diffusion
Under Pressure: Emotional Framing Induces Measurable Behavioral Shifts and Structured Internal Geometry in Small Language ModelsMechanistic interpretability → Activation steering methods
Long-Context Reasoning Through Proxy-Based Chain-of-Thought TuningLLM reasoning → Test-time compute and efficient thinking
FlowLM: Few-Step Language Modeling via Diffusion-to-Flow AdaptationDiffusion models → Discrete and text diffusion
DOTRAG: Retrieval-Time Reasoning Along PathsRetrieval-augmented generation → Knowledge-graph and structured-knowledge RAG
DisastRAG: A Multi-Source Disaster Information Integration and Access System Based on Retrieval-Augmented Large Language ModelsRetrieval-augmented generation → Agentic RAG, agents and memory
CreativityBench: Evaluating Agent Creative Reasoning via Affordance-Based Tool RepurposingLLM reasoning → Reasoning benchmarks and evaluation +1
Delay, Plateau, or Collapse: Evaluating the Impact of Systematic Verification Error on RLVRLLM reasoning → Reinforcement learning for reasoning
Memorization In Stable Diffusion Is Unexpectedly Driven by CLIP EmbeddingsDiffusion models → Diffusion foundations and sampling
eOptShrinkQ: Near-Lossless KV Cache Compression Through Optimal Spectral Denoising and QuantizationEfficient LLM inference and serving → KV cache optimization
Fuzzy Encoding-Decoding to Improve Spiking Q-Learning Performance in Autonomous DrivingEnd-to-end autonomous driving → End-to-end system architecture +1
HalluSAE: Detecting Hallucinations in Large Language Models via Sparse Auto-EncodersMechanistic interpretability → Hallucination detection from model internals
CAGE: Bridging the Accuracy-Aesthetics Gap in Educational Diagrams via Code-Anchored Generative EnhancementDiffusion models → Image generation, text-to-image and editing
Grid2Matrix: Revealing Digital Agnosia in Vision-Language ModelsLLM reasoning → Multimodal reasoning +1
Spectral Edge Dynamics Reveal Functional Modes of LearningMechanistic interpretability → Circuits and mechanistic analysis
DISSECT: Diagnosing Where Vision Ends and Language Priors Begin in Scientific VLMsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
SALLIE: Generation-Free Hidden-State Detection of Jailbreaks and Prompt Injections Across Text and VisionMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness +1
Extending Tabular Denoising Diffusion Probabilistic Models for Time-Series Data GenerationDiffusion models → Flow matching, time series and science
DualDiffusion: A Speculative Decoding Strategy for Masked Diffusion ModelsDiffusion models → Discrete and text diffusion +1
XMark: Reliable Multi-Bit Watermarking for LLM-Generated TextsLLM safety and security → Watermarking generated content and models
From Governance Norms to Enforceable Controls: A Layered Translation Method for Runtime Guardrails in Agentic AILLM agents and tool use → Agent threats, guardrails and governance
Integration of Object Detection and Small VLMs for Construction Safety Hazard IdentificationMultimodal LLMs / vision-language models → Chart understanding and anomaly detection
Vehicle-as-Prompt: A Unified Deep Reinforcement Learning Framework for Heterogeneous Fleet Vehicle Routing ProblemDeep reinforcement learning methods → RL for games and combinatorial optimization
Retrieval-Augmented LLMs for Evidence Localization in Clinical Trial Recruitment from Longitudinal EHR NarrativesRetrieval-augmented generation → Medical and biomedical RAG
Cross-fitted Proximal Learning for Model-Based Reinforcement LearningDeep reinforcement learning methods → RL theory and sample complexity
OrthoFuse: Training-free Riemannian Fusion of Orthogonal Style-Concept Adapters for Diffusion ModelsDiffusion models → Diffusion foundations and sampling
Gradient-Controlled Decoding: A Safety Guardrail for LLMs with Dual-Anchor SteeringLLM safety and security → Jailbreaking
Graph Signal Diffusion Models for Wireless Resource AllocationDiffusion models → Diffusion for compression and wireless
ClawsBench: Evaluating Capability and Safety of LLM Productivity Agents in Simulated WorkspacesLLM agents and tool use → General LLM agent frameworks and benchmarks
Not All Turns Are Equally Hard: Adaptive Thinking Budgets For Efficient Multi-Turn Reasoning in AgentsLLM reasoning → Test-time compute and efficient thinking +1
A Multi-Agent Approach to Validate and Refine LLM-Generated Personalized Math ProblemsLLM agents and tool use → Educational agents and math problem solving
IntentScore: Intent-Conditioned Action Evaluation for Computer-Use AgentsLLM agents and tool use → GUI and mobile agents
EvolveRouter: Co-Evolving Routing and Prompt for Multi-Agent Question AnsweringLLM agents and tool use → Web, search, and research agents
SenseAI: A Human-in-the-Loop Dataset for RLHF-Aligned Financial Sentiment ReasoningLLM reasoning → Reasoning for agents, planning and code +1
How Reasoning Evolves from Post-Training Data: An Empirical Study Using ChessLLM reasoning → Reasoning for agents, planning and code
A Multi-Agent Framework for Automated Exploit Generation with Constraint-Guided Comprehension and ReflectionLLM agents and tool use → Agent threats, guardrails and governance
Offline RL for Adaptive Policy Retrieval in Prior AuthorizationRetrieval-augmented generation → Agentic RAG, agents and memory +2
Watch Before You Answer: Learning from Visually Grounded Post-TrainingMultimodal LLMs / vision-language models → Video understanding with MLLMs
Uncertainty-Guided Latent Diagnostic Trajectory Learning for Sequential Clinical DiagnosisLLM agents and tool use → Medical and scientific domain agents
π2: Structure-Originated Reasoning Data Improves Long-Context Reasoning Ability of Large Language ModelsLLM reasoning → Reasoning benchmarks and evaluation
Vintix II: Decision Pre-Trained Transformer is a Scalable In-Context Reinforcement LearnerDeep reinforcement learning methods → Deep RL algorithms and representations
Simultaneous Dual-View Mammogram Synthesis Using Denoising Diffusion Probabilistic ModelsDiffusion models → Medical imaging and segmentation
Nidus: Externalized Reasoning for AI-Assisted EngineeringLLM reasoning → Reasoning for agents, planning and code
GLANCE: A Global-Local Coordination Multi-Agent Framework for Music-Grounded Non-Linear Video EditingLLM agents and tool use → Multimodal and embodied agents
MMORF: A Multi-agent Framework for Designing Multi-objective Retrosynthesis Planning SystemsLLM agents and tool use → Medical and scientific domain agents
GaussFly: Contrastive Reinforcement Learning for Visuomotor Policies in 3D Gaussian FieldsDeep reinforcement learning methods → RL for robot navigation and UAV control
Video-MME-v2: Towards the Next Stage in Benchmarks for Comprehensive Video UnderstandingWorld models → Latent predictive world models
StarVLA: A Lego-like Codebase for Vision-Language-Action Model DevelopingWorld models → Embodied and robot world models +2
Scaling Coding Agents via Atomic SkillsLLM agents and tool use → Software engineering coding agents
Comparative Characterization of KV Cache Management Strategies for LLM InferenceEfficient LLM inference and serving → KV cache optimization
Early Stopping for Large Reasoning Models via Confidence DynamicsLLM reasoning → Test-time compute and efficient thinking
Rethinking Model Efficiency: Multi-Agent Inference with Large ModelsLLM reasoning → Multimodal reasoning +2
Your Pre-trained Diffusion Model Secretly Knows RestorationDiffusion models → Inverse problems and image restoration
Stratifying Reinforcement Learning with Signal Temporal LogicDeep reinforcement learning methods → Interpretable and logic-guided RL
TriAttention: Efficient Long Reasoning with Trigonometric KV CompressionLLM reasoning → Test-time compute and efficient thinking +1
Comparing Human Oversight Strategies for Computer-Use AgentsLLM agents and tool use → Agent threats, guardrails and governance
Vero: An Open RL Recipe for General Visual ReasoningLLM reasoning → Multimodal reasoning
Analyzing Symbolic Properties for DRL Agents in Systems and NetworkingDeep reinforcement learning methods → RL for energy, networks and operations
A Frame is Worth One Token: Efficient Generative World Modeling with Delta TokensWorld models → Video-generation world models and simulators
HI-MoE: Hierarchical Instance-Conditioned Mixture-of-Experts for Object DetectionMixture-of-experts → Vision mixture-of-experts
Are Latent Reasoning Models Easily Interpretable?LLM reasoning → Test-time compute and efficient thinking
Agentic Federated Learning: The Future of Distributed Training OrchestrationLLM agents and tool use → Agentic AI for wireless networks
Asymmetric Advantage Modulation Calibrates Entropy Dynamics in RLVRLLM reasoning → Reinforcement learning for reasoning
Data Attribution in Adaptive LearningDeep reinforcement learning methods → Deep RL algorithms and representations
Synthetic Sandbox for Training Machine Learning Engineering AgentsLLM agents and tool use → Software engineering coding agents
Beyond the Global Scores: Fine-Grained Token Grounding as a Robust Detector of LVLM HallucinationsLLM reasoning → Hallucination detection and mitigation +1
The Blind Spot of Adaptation: Quantifying and Mitigating Forgetting in Fine-tuned Driving ModelsMultimodal LLMs / vision-language models → Driving VLMs and open-vocabulary detection +1
MemMachine: A Ground-Truth-Preserving Memory System for Personalized AI AgentsRetrieval-augmented generation → Agentic RAG, agents and memory +1
Strengthening Human-Centric Chain-of-Thought Reasoning Integrity in LLMs via a Structured Prompt FrameworkLLM reasoning → Reasoning for agents, planning and code
Full-Duplex-Bench-v3: Benchmarking Tool Use for Full-Duplex Voice Agents Under Real-World DisfluencyLLM agents and tool use → General LLM agent frameworks and benchmarks
InfBaGel: Human-Object-Scene Interaction Generation with Dynamic Perception and Iterative RefinementDiffusion models → Video generation
Do No Harm: Exposing Hidden Vulnerabilities of LLMs via Persona-based Client Simulation Attack in Psychological CounselingLLM safety and security → Safety alignment, refusal and guardrails
Less Detail, Better Answers: Degradation-Driven Prompting for VQAMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
E-VLA: Event-Augmented Vision-Language-Action Model for Dark and Blurred ScenesVision-language-action models → 3D and spatial-aware VLA
Plausibility as Commonsense Reasoning: Humans Succeed, Large Language Models Do notLLM reasoning → Reasoning benchmarks and evaluation
ANX: Protocol-First Design for AI Agent Interaction with a Supporting 3EX Decoupled ArchitectureLLM agents and tool use → General LLM agent frameworks and benchmarks
Selecting Decision-Relevant Concepts in Reinforcement LearningDeep reinforcement learning methods → Interpretable and logic-guided RL
SkillX: Automatically Constructing Skill Knowledge Bases for AgentsLLM agents and tool use → Agent skill learning and evolution
Feasibility-Aware Imitation Learning for Benders DecompositionOffline RL and imitation learning from data → Imitation learning methods
Forgetting to Witness: Efficient Federated Unlearning and Its Visible EvaluationLLM safety and security → Federated unlearning and poisoning
Cog-DRIFT: Exploration on Adaptively Reformulated Instances Enables Learning from Hard Reasoning ProblemsLLM reasoning → Reinforcement learning for reasoning
Your Agent, Their Asset: A Real-World Safety Analysis of OpenClawLLM agents and tool use → Agent threats, guardrails and governance
Undetectable Conversations Between AI Agents via Pseudorandom Noise-Resilient Key ExchangeLLM agents and tool use → Agent threats, guardrails and governance
Discovering Failure Modes in Vision-Language Models using RLLLM reasoning → Multimodal reasoning +1
Metaphors We Compute By: A Computational Audit of Cultural Translation vs. Thinking in LLMsLLM reasoning → Reasoning benchmarks and evaluation
Don't Waste Bits! Adaptive KV-Cache Quantization for Lightweight On-Device LLMsEfficient LLM inference and serving → KV cache quantization
What Makes Good Multilingual Reasoning? Disentangling Traces with Measurable FeaturesLLM reasoning → Reasoning benchmarks and evaluation
OpenWorldLib: A Unified Codebase and Definition of Advanced World ModelsWorld models → World models for language agents
On the "Causality" Step in Policy Gradient Derivations: A Pedagogical Reconciliation of Full Return and Reward-to-GoDeep reinforcement learning methods → Deep RL algorithms and representations
An AI Teaching Assistant for Motion Picture EngineeringRetrieval-augmented generation → Agentic RAG, agents and memory
ROSClaw: A Hierarchical Semantic-Physical Framework for Heterogeneous Multi-Agent CollaborationLLM reasoning → Spatial and embodied reasoning +1
Anticipatory Reinforcement Learning: From Generative Path-Laws to Distributional Value FunctionsDeep reinforcement learning methods → RL theory and sample complexity
Springdrift: An Auditable Persistent Runtime for LLM Agents with Case-Based Memory, Normative Safety, and Ambient Self-PerceptionLLM agents and tool use → Agent memory systems
Search, Do not Guess: Teaching Small Language Models to Be Effective Search AgentsLLM agents and tool use → Web, search, and research agents
From Curiosity to Caution: Mitigating Reward Hacking for Best-of-N with PessimismLLM reasoning → Reinforcement learning for reasoning +1
Training-Free Refinement of Flow Matching with Divergence-based SamplingDiffusion models → Flow matching, time series and science
Multimodal Backdoor Attack on VLMs for Autonomous Driving via Graffiti and Cross-Lingual TriggersMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness +2
Beyond Semantics: Uncovering the Physics of Fakes via Universal Physical Descriptors for Cross-Modal Synthetic DetectionMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness
Ruling Out to Rule In: Contrastive Hypothesis Retrieval for Medical Question AnsweringRetrieval-augmented generation → Medical and biomedical RAG
Firebolt-VL: Efficient Vision-Language Understanding with Cross-Modality ModulationMultimodal LLMs / vision-language models → Visual token pruning and efficiency

All 391