Papers

New on arXiv

Sun 4 Oct 2026

388 papers

Inference-Time Projection for Physically Valid Biomolecular Diffusion ModelsDiffusion models → Flow matching, time series and science
Investigating Model Compression for Neural Machine Translation in the Biomedical DomainKnowledge distillation → Knowledge distillation for LLMs
Artemis: Geometry-Grounded Multi-Agent Driving World Models with Shared 3D State and Progressive Memory UpdateWorld models → Video-generation world models and simulators
Identifiable World Models from Pretrained Diffusion RepresentationsWorld models → Latent predictive world models +1
Beyond Refusal Patterns: Safe-Role Internalization for Robust and Generalizable LLM Safety AlignmentLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
When to Rethink: Learning Multi-Perspective Self-Verification for Vision-Language ModelsLLM reasoning → Multimodal reasoning +1
Which Image Property Carries the Jailbreak? A Controlled Dissection of Image-to-Text JailbreaksLLM safety and security → Jailbreaking
Where Does the Audio Jailbreak Live? A Controlled Frequency-Depth Audit of AdvWave-P on Qwen2-AudioLLM safety and security → Audio jailbreaks and speech backdoors
Topology-Consistent Task Planning over Cellular Workflow Complexes for LLM-based AgentsLLM agents and tool use → Tool learning and function calling
Should We Skip Diffusion?Diffusion models → Diffusion foundations and sampling
Mask-Guided KV Cache Eviction in Block Diffusion Language ModelsDiffusion models → Discrete and text diffusion +1
DART-ES: Difficulty-Aware Reweighting and Targeted Replay for Fine-Tuning LLMs with Evolution StrategiesLLM reasoning → Test-time compute and efficient thinking
EPOCH: Reliable Discovery through Evidence-Governed SearchLLM agents and tool use → Web, search, and research agents
Visual-Invariance-Augmented Feature Optimal Alignment for Transferable Adversarial Attacks against Closed-Source MLLMsMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness
Atomic Visual Entailment: Enhancing Zero-Shot Vision-Language Reasoning through Atomic Fact Decomposition and Learned SelectionLLM reasoning → Multimodal reasoning +1
DREAM: Dynamic Resolution Assignment For Multimodal Multi-agent DebateLLM reasoning → Multimodal reasoning +1
Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio GenerationDiffusion models → Audio, speech and music
Your Unlearning Gives You Away: Identifying Erased Concepts in Diffusion ModelsDiffusion models → Robustness, privacy and adversarial aspects +1
AgentDoxx: Agentic Re-identification of Anonymized Text with Web SearchLLM agents and tool use → Agent threats, guardrails and governance
Expanding LLM ReasoningLLM reasoning → Test-time compute and efficient thinking
Factoriax: A GPU-Accelerated Factorio-Style Simulator for Reinforcement LearningDeep reinforcement learning methods → RL for games and combinatorial optimization
LifeLong Digital Twin: A Unified Modeling Paradigm and Agent Harness for Event-Driven Lifelong Health State TrajectoriesLLM agents and tool use → Medical and scientific domain agents
G-CARB: Graph-Localized Conformal Agent Risk Budget for Compositional HarmLLM agents and tool use → Agent threats, guardrails and governance
When Low Prediction Error Misleads Planning: Diagnosing Representation, Dynamics, and Decision Failures in Latent World ModelsWorld models → World models for language agents
Don't Judge an LLM Only by Its Activations: Discovering Suppressed Safety Features via Counterfactual Activation PotentialLLM safety and security → Safety alignment, refusal and guardrails +1
SALUS: Automated Auditing of NL-to-SQL Benchmarks through Weak Supervision of Multi-Agent OutputLLM agents and tool use → Web, search, and research agents
LiFT: Loop Flow TransformersDiffusion models → Flow matching, time series and science
When Does Retrieval Help? A Study of In-Context Adaptation in Vision-Language-Action ModelsVision-language-action models → Generalist VLA models and benchmarks
Universality and Convergence of Generative FlowsDiffusion models → Flow matching, time series and science
Universal Test-Time TrainingMixture-of-experts → Efficient MoE training and inference systems
FLEX-WAM: Flexible Block-Causal World-Action Models for Long-Horizon Imagination and PlanningWorld models → Embodied and robot world models
CodeForge-MA: Execution-Verified Multi-Agent Learning with Language-Conditioned LoRA for Multilingual Code GenerationLLM agents and tool use → Software engineering coding agents
Hierarchical Reinforcement Learning with Stable Temporal Abstraction for Language Model AgentsLLM agents and tool use → Reinforcement learning for LLM agents
Hallucination Across the Reasoning Lifecycle: Interface Visibility, Causal Evidence, and Release Control in Large Reasoning ModelsLLM reasoning → Hallucination detection and mitigation
AI Safety via Debate is Compromised by Cognitive BiasesLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
Measuring and Reducing Cross-Vendor Mismatch in Language ModelsMixture-of-experts → MoE large language models
The Poisoned Conversation: Privacy-Leaking Watermarks in Unified Multimodal ModelsLLM safety and security → Watermarking generated content and models
Groupwise Distortion Guarantees for Preference-Based AlignmentLLM alignment / RLHF / preference optimisation → Reward modeling and RLHF
Hierarchical Time-aware Bootstrapping for Off-Policy Subgoal Value LearningDeep reinforcement learning methods → Deep RL algorithms and representations
TeleTune: Evolving Agent Skills From Offline TelemetryLLM agents and tool use → Agent skill learning and evolution
OpticalRec: Unified Optical Vision-Language Representation for Multimodal RecommendationMultimodal LLMs / vision-language models → Document understanding and multimodal RAG
EvoMem-VLA: State-Evolution Memory for Long-Horizon Robot ManipulationVision-language-action models → Reasoning and chain-of-thought in VLA
Render to Reason: Novel-View Semantic Prediction Improves Spatial Understanding in VLMsLLM reasoning → Spatial and embodied reasoning +1
RMMBench: A Comprehensive Benchmark for Robotic Mobile ManipulationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
A Strong Baseline for Evaluating Vision Encoders in Multimodal Large Language ModelsMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
CleanMDM: Clean Motion Diffusion Model for Multimodal Motion CleanupDiffusion models → Video generation
BeliefGraph-JEPA: Structured Latent World Models for Action-Conditioned Time SeriesWorld models → Latent predictive world models
No Concept Escapes the Audit: Auditing-Aware Unlearning for Verifiable Concept Erasure in Diffusion ModelsDiffusion models → Robustness, privacy and adversarial aspects +1
CASE: Cost-Aware Stopping for Efficient Long-Video AgentsLLM agents and tool use → Agent memory systems
Efficient Graph Generation via Direct Prediction and Flow MatchingDiffusion models → Flow matching, time series and science
Towards Unbiased On-Policy Distillation for Block Diffusion Language ModelsDiffusion models → Discrete and text diffusion
Robust Ensemble Guidance for Scientific Inverse ProblemsDiffusion models → Inverse problems and image restoration
EnGRICH: Enhancing Generative Reward Modeling with Critiques from HumansLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
AIProver: Agentic Auto-Formalization of Mathematical Research via Certificate-Driven Evolving HarnessLLM agents and tool use → Educational agents and math problem solving
Optimal Control with Learned Critics under Unmodeled State DependenciesDeep reinforcement learning methods → RL theory and sample complexity
Learning Conditional Source Distribution via Flow Reversal for Temporal Flow MatchingDiffusion models → Diffusion and flow policies for robotics
Reflections and Fragments: Securing LLMs Against Sequential Mosaic AttacksLLM safety and security → Prompt injection and agent security
IRSTD-Agent: Agentic Infrared Small Target Detection via Zoom-Guided Interaction LearningMultimodal LLMs / vision-language models → Embodied and 3D vision-language models +1
On Semi-Markov Suboptimality in Hierarchical Reinforcement LearningDeep reinforcement learning methods → Deep RL algorithms and representations
Diffusion Transformers are Provably Optimal In-context GeneratorsDiffusion models → Diffusion and flow policies for robotics
Shadow Feature Refinement Network: Progressive Feature Refinement based on Knowledge Distillation for Effective Shadow RemovalKnowledge distillation → Detection and dense prediction distillation
When Does Longer Reasoning Help? Predicting Mathematical Reasoning Through Discovery and ExecutionLLM reasoning → Mathematical reasoning
RubricArmor: Adversarial Evolution Improves LLM-Based Rubric GenerationLLM alignment / RLHF / preference optimisation → Reward hacking and misspecification
ASCENT: Online Test-Time Training of Long-Horizon Agents via Self-Distillation of Verified ExperienceLLM agents and tool use → Agent skill learning and evolution +1
MESH-Harness: Self-Improving Agent Harnesses via Bandit-Guided Compositional EvolutionLLM agents and tool use → Software engineering coding agents +1
Readable Before Actionable: Causal Tracing of Indirect Prompt InjectionLLM agents and tool use → Agent threats, guardrails and governance +1
Fusion is the New Mutation: Bandit-Guided Evolution on Workflow GraphsLLM agents and tool use → Software engineering coding agents
Red-TTT: Test-Time Training for Automated Jailbreaking Large Language ModelsLLM safety and security → Jailbreaking
When and What to Prune? Stage-Aware Visual Token Pruning for Efficient VLAMultimodal LLMs / vision-language models → Visual token pruning and efficiency +1
Who Is Your Agent Serving? Provider-Side Indirect Prompt Injection in Proactive AgentsLLM safety and security → Prompt injection and agent security
Learning from imperfect teachers for low-resource acoustic generalizationKnowledge distillation → Speech and audio distillation
Hybrid-Basis Feature Forecasting for Diffusion Sampling AccelerationDiffusion models → Flow matching, time series and science
MGPO: Manifold-Guided Diffusion Alignment for Task-Aware Dataset DistillationDiffusion models → Diffusion and flow policies for robotics +1
templar: agentic induction and evolution of standardized radiology reporting templates from large-scale clinical corporaLLM agents and tool use → Medical and scientific domain agents
StateWise: Diagnosing and Repairing Persistent Operational State Before Agent ActionsLLM reasoning → Reasoning for agents, planning and code +1
Pythia: Toward Foundation World Models for Multimodal Time SeriesWorld models → Latent predictive world models
PixReenact: Pixel-Conditioned Causal Video Diffusion for Streaming Head-Avatar ReenactmentDiffusion models → Video generation
Vela: Scaling Vision-Language-Action Models with Adaptive Action Curve ParametrizationVision-language-action models → Generalist VLA models and benchmarks
Look Before You Leap: Thermodynamic Arbitration of Parametric and Non-Parametric Knowledge in LLM Agents via Self-Regulating Memory ArchitecturesLLM agents and tool use → Agent memory systems
Safe Context Switching for Agents in the Wild: Mitigating Subspace Interference via Orthogonal AdaptationLLM reasoning → Reasoning for agents, planning and code +3
Learning without Overwriting: A Theory of Self-Distillation and Supervised Fine-Tuning in Continual ReasoningLLM reasoning → Test-time compute and efficient thinking +1
Cross-Time Directional Selection in Diffusion SamplingDiffusion models → Diffusion foundations and sampling
Same Predictions, Different Harms: Causal Auditing of Patient World ModelsWorld models → Medical and surgical world models
Long-MDR: Long-Context Reinforcement Learning for Multimodal Deep-Research AgentsLLM reasoning → Reinforcement learning for reasoning +1
Recurrent Latent Visual Search for GUI GroundingMultimodal LLMs / vision-language models → Embodied and 3D vision-language models +1
AECG: Asymmetric Experience Consolidation and Governance In Multi-Agent SystemsLLM agents and tool use → Agent memory systems
A Safe Action Is Not Enough: Feasible-Future Decoding for Vision-Language-Action PoliciesVision-language-action models → Failure recovery and robustness
Blocking at the Boundary: Auditing Long-Horizon Agents against Staged Prompt InjectionLLM safety and security → Prompt injection and agent security
Memadapter: Counterfactual Adaptation Against Memory-induced SycophancyLLM agents and tool use → Agent memory systems
Arithmetic Actor Heads and Training Stabilization for Out-of-Distribution Reinforcement LearningDeep reinforcement learning methods → Deep RL algorithms and representations
RoMod: Temporal Routing Modulation via Mixture-of-Experts for Video Anomaly DetectionMixture-of-experts → Vision mixture-of-experts
Representation--Behavior Alignment for Explainable Weakly-Supervised Video Anomaly DetectionMultimodal LLMs / vision-language models → Video understanding with MLLMs
Bayesian Entropy-based Reordering for Calibrated Diffusion Language ModelsDiffusion models → Discrete and text diffusion
Memory Canonicalization: A Framework and Benchmark for Cross-Model Drift in Persistent LLM MemoryLLM agents and tool use → Agent memory systems
LexiHorizon: Stabilizing Reinforcement Learning for Long-Horizon Deep SearchLLM reasoning → Knowledge- and tool-augmented reasoning
LoopMoEVR: Loop-Based Degradation-Aware Mixture-of-Experts for Unified UHD Video RestorationMixture-of-experts → Vision mixture-of-experts
Best-of-N Guidance for Test-time Diffusion AlignmentDiffusion models → Diffusion foundations and sampling +1
SearchJev: A Fast and Calibrated System-1 Model for Search AgentsLLM agents and tool use → Web, search, and research agents
SharpDraft: Accelerating Long-Context Speculative Decoding with Cardinality-Aware Query ScalingEfficient LLM inference and serving → Speculative decoding
ReMAP: Restoring the Perceptual Cycle with Reasoning-Time Latent Visual MemoryLLM reasoning → Multimodal reasoning +1

All 388