Papers

New on arXiv

Wed 7 Jan 2026

401 papers

PILOT: Planning via Internalized Latent Optimization Trajectories for Large Language ModelsLLM reasoning → Test-time compute and efficient thinking
Towards Latent Diffusion Suitable For TextDiffusion models → Discrete and text diffusion
MALTopic: Multi-Agent LLM Topic Modeling FrameworkLLM agents and tool use → Web, search, and research agents
MixDPO: Modeling Preference Strength for Pluralistic AlignmentLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
TIR-Flow: Active Video Search and Reasoning with Frozen VLMsLLM reasoning → Multimodal reasoning
Think Bright, Diffuse Nice: Enhancing T2I-ICL via Inductive-Bias Hint Instruction and Query Contrastive DecodingLLM reasoning → Multimodal reasoning
What Users Leave Unsaid: Under-Specified Queries Limit Vision-Language ModelsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Forget-It-All: Multi-Concept Machine Unlearning via Concept-Aware Neuron MaskingDiffusion models → Robustness, privacy and adversarial aspects +1
UniDrive-WM: Unified Understanding, Planning and Generation World Model for Autonomous DrivingWorld models → World models for autonomous driving +1
Merging Triggers, Breaking Backdoors: Defensive Poisoning for Instruction-Tuned Language ModelsLLM safety and security → Backdoor and data poisoning attacks
Addressing Overthinking in Large Vision-Language Models via Gated Perception-Reasoning OptimizationLLM reasoning → Multimodal reasoning +1
Improving and Accelerating Offline RL in Large Discrete Action Spaces with Structured Policy InitializationOffline RL and imitation learning from data → Offline RL algorithms
XGrammar-2: Dynamic and Efficient Structured Generation Engine for Agentic LLMsLLM agents and tool use → Software engineering coding agents
Gavel: Agent Meets Checklist for Evaluating LLMs on Long-Context Legal SummarizationLLM agents and tool use → Social simulation and multi-agent debate
Distribution-Guided and Constrained Quantum Machine UnlearningLLM safety and security → Machine unlearning
Rate or Fate? RLV^varepsilonR: Reinforcement Learning with Verifiable Noisy RewardsLLM alignment / RLHF / preference optimisation → Process reward models for reasoning +1
3D-Agent:Tri-Modal Multi-Agent Collaboration for Scalable 3D Object AnnotationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models +1
Interpreting Transformers Through Attention Head InterventionMechanistic interpretability → Circuits and mechanistic analysis
Enhanced-FQL(λ), an Efficient and Interpretable RL with novel Fuzzy Eligibility Traces and Segmented Experience ReplayDeep reinforcement learning methods → RL theory and sample complexity
Safety Is Not Universal: The Selective Safety Trap in LLM AlignmentLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
Few-Shot LoRA Adaptation of a Flow-Matching Foundation Model for Cross-Spectral Object DetectionDiffusion models → Inverse problems and image restoration
Disco-RAG: Discourse-Aware Retrieval-Augmented GenerationRetrieval-augmented generation → RAG pipeline design and evaluation
Survival Dynamics of Neural and Programmatic Policies in Evolutionary Reinforcement LearningDeep reinforcement learning methods → RL for games and combinatorial optimization
PackCache: A Training-Free Acceleration Method for Unified Autoregressive Video Generation via Compact KV-CacheEfficient LLM inference and serving → Fast autoregressive visual generation
ReHyAt: Recurrent Hybrid Attention for Video Diffusion TransformersDiffusion models → Video generation
Autonomous Reasoning for Spacecraft Control: A Large Language Model Framework with Group Relative Policy OptimizationLLM reasoning → Reinforcement learning for reasoning
Beyond Binary Preference: Aligning Diffusion Models to Fine-grained Criteria by Decoupling AttributesDiffusion models → Diffusion and flow policies for robotics +1
Online Action-Stacking Improves Reinforcement Learning Performance for Air Traffic ControlDeep reinforcement learning methods → RL for robot navigation and UAV control
LEGATO: Good Identity Unlearning Is ContinuousLLM safety and security → Machine unlearning
From Domains to Instances: Dual-Granularity Data Synthesis for LLM UnlearningSynthetic data and self-improvement for LLMs → Self-rewarding alignment and self-training +1
Shadow Unlearning: A Neuro-Semantic Approach to Fidelity-Preserving Faceless Forgetting in LLMsLLM safety and security → Machine unlearning
Correcting Autonomous Driving Object Detection Misclassifications with Automated Commonsense ReasoningLLM reasoning → Chain-of-thought reasoning for driving
Replacing Tunable Parameters in Weather and Climate Models with State-Dependent Functions using Reinforcement LearningDeep reinforcement learning methods → RL for energy, networks and operations
State Backdoor: Towards Stealthy Real-world Poisoning Attack on Vision-Language-Action Model in State SpaceVision-language-action models → Safety and adversarial attacks +1
MemKD: Memory-Discrepancy Knowledge Distillation for Efficient Time Series ClassificationKnowledge distillation → General knowledge distillation methods
Learning to Reason: Temporal Saliency Distillation for Interpretable Knowledge TransferKnowledge distillation → General knowledge distillation methods
Safety-Utility Conflicts Are Not Global: Surgical Alignment via Head-Level DiagnosisLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
Inhibitory Attacks on Backdoor-based Fingerprinting for Large Language ModelsLLM safety and security → Watermarking generated content and models
Towards a Mechanistic Understanding of Propositional Logical Reasoning in Large Language ModelsLLM reasoning → Reasoning benchmarks and evaluation
Agentic Rubrics as Contextual Verifiers for SWE AgentsLLM reasoning → Reasoning for agents, planning and code +1
Agent Drift: Quantifying Behavioral Degradation in Multi-Agent LLM Systems Over Extended InteractionsLLM agents and tool use → General LLM agent frameworks and benchmarks
Diffusion-DRF: Free, Rich, and Differentiable Reward for Video Diffusion Fine-TuningDiffusion models → Video generation +1
Wow, wo, val! A Comprehensive Embodied World Model Evaluation Turing TestWorld models → World models for navigation
ContextFocus: Activation Steering for Contextual Faithfulness in Large Language ModelsMechanistic interpretability → Activation steering methods
InfiniteWeb: Scalable Web Environment Synthesis for GUI Agent TrainingLLM agents and tool use → Web, search, and research agents
GeoReason: Aligning Thinking And Answering In Remote Sensing Vision-Language Models Via Logical Consistency Reinforcement LearningLLM reasoning → Spatial and embodied reasoning +1
SearchAttack: Red-Teaming LLMs against Knowledge-to-Action Threats under Online Web SearchLLM safety and security → Prompt injection and agent security
Gen3R: 3D Scene Generation Meets Feed-Forward ReconstructionDiffusion models → 3D, point cloud and scene generation
KDCM: Reducing Hallucination in LLMs through Explicit Reasoning StructuresLLM reasoning → Hallucination detection and mitigation
Cells on Autopilot: Adaptive Cell (Re)Selection via Reinforcement LearningDeep reinforcement learning methods → RL for energy, networks and operations
Analyzing Reasoning Consistency in Large Multimodal Models under Cross-Modal ConflictsLLM reasoning → Multimodal reasoning +1
Mind the Generative Details: Direct Localized Detail Preference Optimization for Video Diffusion ModelsDiffusion models → Video generation +1
CLAP: Contrastive Latent Action Pretraining for Learning Vision-Language-Action Models from Human VideosVision-language-action models → Human-video and latent-action pretraining
Bridging the Discrete-Continuous Gap: Unified Multimodal Generation via Coupled Manifold Discrete Absorbing DiffusionDiffusion models → Image generation, text-to-image and editing
Modular Prompt Optimization: Optimizing Structured Prompts with Section-Local Textual GradientsLLM reasoning → Prompting and chain-of-thought reasoning
LinkD: AutoRegressive Diffusion Model for Mechanical Linkage SynthesisDiffusion models → Diffusion and flow policies for robotics
Stable Language Guidance for Vision-Language-Action ModelsVision-language-action models → Failure recovery and robustness
When Helpers Become Hazards: A Benchmark for Analyzing Multimodal LLM-Powered Safety in Daily LifeMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness
MobileDreamer: Generative Sketch World Model for GUI AgentWorld models → World model security and GUI agents +1
HoneyTrap: Deceiving Large Language Model Attackers to Honeypot Traps with Resilient Multi-Agent DefenseLLM agents and tool use → Agent threats, guardrails and governance +1
Thinking with Frames: Generative Video Distortion Evaluation via Frame Reward ModelLLM alignment / RLHF / preference optimisation → Diffusion and multimodal model alignment
A Scheduling Framework for Efficient MoE Inference on Edge GPU-NDP SystemsMixture-of-experts → Efficient MoE training and inference systems
RADAR: Retrieval-Augmented Detector with Adversarial Refinement for Adaptive LLM-Generated Fake News DetectionRetrieval-augmented generation → Privacy and security of RAG
SoK: Privacy Risks and Mitigations in Retrieval-Augmented Generation SystemsRetrieval-augmented generation → Privacy and security of RAG
On-Device Deep Reinforcement Learning for Decentralized Task Offloading Performance trade-offs in the training processDeep reinforcement learning methods → RL for energy, networks and operations
Anti-Length Shift: Dynamic Outlier Truncation for Training Efficient Reasoning ModelsLLM reasoning → Test-time compute and efficient thinking
CoINS: Counterfactual Interactive Navigation via Skill-Aware VLMMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Trade-R1: Bridging Verifiable Rewards to Stochastic Environments via Process-Level Reasoning VerificationLLM reasoning → Reasoning for agents, planning and code
Large-Scale Aspect-Based Sentiment Analysis with Reasoning-Infused LLMsLLM reasoning → Audio-language model reasoning
FocusUI: Efficient UI Grounding via Position-Preserving Visual Token SelectionMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Doc-PP: Document Policy Preservation Benchmark for Large Vision-Language ModelsMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness
HemBLIP: A Vision-Language Model for Interpretable Leukemia Cell Morphology AnalysisMultimodal LLMs / vision-language models → Medical vision-language models
Decide Then Retrieve: A Training-Free Framework with Uncertainty-Guided Triggering and Dual-Path RetrievalRetrieval-augmented generation → RAG robustness and adaptive retrieval
Current Agents Fail to Leverage World Model as Tool for ForesightWorld models → World models for language agents +1
Towards Safe Autonomous Driving: A Real-Time Motion Planning Algorithm on Embedded HardwareEnd-to-end autonomous driving → Adversarial scenarios and safety testing
All-Quadrant Bounded Clipping GRPO: Closing the Unbounded Blind Spot for Stable and Generalizable TrainingLLM reasoning → Reinforcement learning for reasoning
Spectral Manifold Regularization for Stable and Modular Routing in Deep MoE ArchitecturesMixture-of-experts → Expert routing strategies
Feature-Aware One-Shot Federated Learning via Hierarchical Token SequencesKnowledge distillation → Federated, robust and private distillation
Staged Voxel-Level Deep Reinforcement Learning for 3D Medical Image Segmentation with Noisy AnnotationsDeep reinforcement learning methods → RL for clinical treatment and surgery
Atlas: Orchestrating Heterogeneous Models and Tools for Multi-Domain Complex ReasoningLLM reasoning → Reasoning for agents, planning and code +1
What Matters For Safety Alignment?LLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
When Numbers Start Talking: Implicit Numerical Coordination Among LLM-Based AgentsLLM agents and tool use → Social simulation and multi-agent debate
Step Potential Advantage Estimation: Harnessing Intermediate Confidence and Correctness for Efficient Mathematical ReasoningLLM reasoning → Reinforcement learning for reasoning
ROI-Reasoning: Rational Optimization for Inference via Pre-Computation Meta-CognitionLLM reasoning → Test-time compute and efficient thinking
VietMed-MCQ: A Consistency-Filtered Data Synthesis Framework for Vietnamese Traditional Medicine EvaluationSynthetic data and self-improvement for LLMs → Medical synthetic data and agents
NeoAMT: Neologism-Aware Agentic Machine Translation with Reinforcement LearningLLM agents and tool use → General LLM agent frameworks and benchmarks
Membox: Weaving Topic Continuity into Long-Range Memory for LLM AgentsLLM agents and tool use → Agent memory systems
PointWorld: Scaling 3D World Models for In-The-Wild Robotic ManipulationWorld models → Embodied and robot world models
EntroCoT: Enhancing Chain-of-Thought via Adaptive Entropy-Guided SegmentationLLM reasoning → Test-time compute and efficient thinking
Bridging OLAP and RAG: A Multidimensional Approach to the Design of Corpus PartitioningRetrieval-augmented generation → RAG pipeline design and evaluation
Whose Facts Win? LLM Source Preferences under Knowledge ConflictsRetrieval-augmented generation → RAG robustness and adaptive retrieval
O-Researcher: An Open Ended Deep Research Model via Multi-Agent Distillation and Agentic RLLLM agents and tool use → Web, search, and research agents
From Laboratory to Real-World Applications: Benchmarking Agentic Code Reasoning at the Repository LevelLLM reasoning → Reasoning for agents, planning and code +1
ETR: Outcome-Guided Elastic Trust Regions for Policy OptimizationLLM reasoning → Reinforcement learning for reasoning
MIND: From Passive Mimicry to Active Reasoning through Capability-Aware Multi-Perspective CoT DistillationLLM reasoning → Prompting and chain-of-thought reasoning
R3L: Reflect-then-Retry Reinforcement Learning with Language-Guided Exploration, Pivotal Credit, and Positive AmplificationLLM reasoning → Reinforcement learning for reasoning +2
BREATH-VL: Vision-Language-Guided 6-DoF Bronchoscopy Localization via Semantic-Geometric FusionMultimodal LLMs / vision-language models → Medical vision-language models
Investigating Knowledge Distillation Through Neural Networks for Protein Binding Affinity PredictionKnowledge distillation → Graph neural network distillation
TreeAdv: Tree-Structured Advantage Redistribution for Group-Based RLLLM reasoning → Reinforcement learning for reasoning +1
Inference Attacks Against Graph Generative Diffusion ModelsDiffusion models → Robustness, privacy and adversarial aspects

All 401