Papers

New on arXiv

Sun 6 Sep 2026

213 papers

Mitigating LLM Over-Refusal via Dynamic Semantic Routing CalibrationLLM safety and security → Safety alignment, refusal and guardrails
Teaching a Moving Student: Rethinking the Curriculum of On-Policy DistillationKnowledge distillation → On-policy and self-distillation for LLMs
RS-Claw-Evolution: Environment-Feedback-Driven Evolution for Lightweight Remote Sensing Agents in Long-Horizon TasksLLM agents and tool use → Multimodal and embodied agents
DIPLOMAT: Dialogue-Span-Aware Direct Preference Optimization for Polite Persuasive Workplace Negotiation DialoguesLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
Deep Persona: A Psychologically Grounded Architecture and Evaluation Framework for Role-Playing Agents and SimulationsLLM agents and tool use → Social simulation and multi-agent debate
Teacher Should Think Ahead: Adaptive Continuations for Reliable On-Policy DistillationLLM reasoning → Prompting and chain-of-thought reasoning +1
From Process Loss to Assembly Bonus: Human-Grounded Diagnosis of Multi-Agent LLM CollaborationLLM agents and tool use → Social simulation and multi-agent debate
TryOnReward: Learning Foveated Consistency for Reinforcement Fine-Tuning of Virtual Try-OnLLM alignment / RLHF / preference optimisation → Diffusion and multimodal model alignment
Sampling headroom is not selection gain: a compute-value audit of test-time scaling for video world modelsWorld models → Video-generation world models and simulators +1
The Double Measurement Confound in Agent Benchmarks: De-Scaffolding, Ground-Truth Scoring, and Reliability Beyond the MeanLLM agents and tool use → General LLM agent frameworks and benchmarks
AgentHijack: Visual Patch Attacks on Multimodal Computer-Use AgentsLLM agents and tool use → Agent threats, guardrails and governance
Identifying Habit, Physics, and Nuisance in Robot World ModelsWorld models → Embodied and robot world models
AutoLexSteer: Automatic Contrast Construction for Lexical Activation SteeringMechanistic interpretability → Activation steering methods
Feature Superposition in Neural Networks: From Theory to PracticeMechanistic interpretability → Sparse autoencoder training and evaluation
Characterizing Contention-Induced Reliability Collapse in KV-Cache Timing Side Channels for Multi-Tenant LLM ServingEfficient LLM inference and serving → KV cache optimization
ContextFlow: In-Context Flow Matching for Robot ManipulationDiffusion models → Diffusion and flow policies for robotics +1
BinauralVAE: Spatial Audio Reconstruction For World ModelsWorld models → Video-generation world models and simulators
Skynet: Workflow-Level Anomaly Detection for Agentic AI via Semantic and Structural ModelingLLM agents and tool use → Software engineering coding agents
NormViz: A Benchmark and Framework for Grounding Multimodal Reasoning in Global CulturesLLM reasoning → Multimodal reasoning
Diagnosing and Dynamically Filtering Occupancy World Models for Active MappingWorld models → World models for autonomous driving
Typed Federated Artifacts for the Agentic Web:Sharing Tool-Routing Knowledge Across Frozen,Heterogeneous LLM AgentsLLM agents and tool use → Agent threats, guardrails and governance
Improving Proficiency and Efficiency of Android GUI Agents via Self-Generating Tool ActionsLLM agents and tool use → GUI and mobile agents
When Retain Constraints Conflict: Mitigating Forget-Retain Interference in Tabular DataLLM safety and security → Machine unlearning
AURA-Eval: Evaluation Framework for Acting Under Risk Awareness in LLM Agent TrajectoriesLLM agents and tool use → Agent threats, guardrails and governance
Shortcutting the Fix: Identifying and Categorizing Agentic Exploits in Software Engineering BenchmarksLLM agents and tool use → Software engineering coding agents
DrugReason: Dynamic Multi-View Reasoning over Knowledge Graph and Language Evidence for Drug RepurposingLLM reasoning → Clinical and biomedical applications
Ordinary, Reasonable Chatbots: Do AI Models Track Human Legal Judgments?LLM reasoning → Reasoning for agents, planning and code
LATS: Levy Adaptive Tree Sampling for Feedback-Driven Diverse Target DiscoveryDiffusion models → Diffusion foundations and sampling
Agentic Visual Generation: From Generative Models to Agentic ControlLLM agents and tool use → Multimodal and embodied agents
Reason Through the Latent! Making Latent Visual Reasoning NecessaryLLM reasoning → Multimodal reasoning
Uni-Light: An Ultra-Lightweight Framework via Uncertainty-Aware Knowledge Distillation for Brain Tumour SegmentationKnowledge distillation → Distillation for medical imaging
RoLA: Rotary-Positioned Low-Rank Linear Attention for Efficient Diffusion TransformersDiffusion models → Video generation
A Trustworthy Watermarking Framework for LLM-Generated Food Safety ContentLLM safety and security → Watermarking generated content and models
Counterfactual Tests for Measuring Chain-of-Thought Faithfulness in Visual Language ModelsLLM reasoning → Multimodal reasoning +1
PARSER: Read in Parallel, Reason in Depth for Long-Context LLM AgentsLLM agents and tool use → Agent memory systems
Behavioral Cloning Outperforms Entropy-Regularized RL: Critic-Driven Failure of Actor-Critic Methods on Adaptive Tumor TreatmentOffline RL and imitation learning from data → Imitation learning methods +1
TransNormal-2: Geometry-Grounded Rectified Flow with Edge-Aware Decoding for Precise Normal EstimationDiffusion models → 3D, point cloud and scene generation
ECOKV: Geometry-Aware KV Cache Eviction via Complementary Diversity MetricsMultimodal LLMs / vision-language models → Visual token pruning and efficiency +1
SwiftExplorer: Training-free Diffusion Model Alignment with Swift Diversity ExplorationDiffusion models → Diffusion foundations and sampling +1
Inducing Emergent Misalignment from Reward Hacks with Iterative DPOLLM alignment / RLHF / preference optimisation → Reward hacking and misspecification
MemCorr-DP: Counterfactual Correspondence Conditioning for a Diffusion Policy Guided by a ReferenceDiffusion models → Diffusion and flow policies for robotics
MechAudit-40: White-Box Auditing across 40 LLM Attack MechanismsLLM safety and security → Prompt injection and agent security
Learning to Use Imagination: Progress-Conditioned Future Utilization for World Action ModelsWorld models → Embodied and robot world models +1
A Translational Note on AI Safety EvaluationLLM safety and security → Prompt injection and agent security
MARBO: Relational Belief Grounding for LLM Agents in Social Deduction GamesLLM agents and tool use → Reinforcement learning for LLM agents
EStream: Fast and Memory-Efficient MoE Prefill through Expert Virtualization on Mobile NPUsMixture-of-experts → Efficient MoE training and inference systems
SRD-GUARD: A Defense Framework of LLMs via Semantic Rewriting and Joint Multi-Model Scoring for Latent Intent ExposureLLM safety and security → Jailbreaking
3DHarnessBench: Probing Agentic 3D-to-Code Capabilities of Frontier Vision-Language ModelsMultimodal LLMs / vision-language models → Embodied and 3D vision-language models +1
Selective Knowledge Control for Continual GUI Agent Learning over Application StreamsLLM agents and tool use → GUI and mobile agents
VLA-Corrector: Stage-Aware Observable State Understanding for Prompt-Based Closed-Loop Recovery of Vision-Language-Action PoliciesVision-language-action models → Failure recovery and robustness
CAM: Question Answering on Entity-Centric Videos with Continuous Extraction and Adaptive QueryingMultimodal LLMs / vision-language models → Video understanding with MLLMs
CAPMAS: Capability-Based Delegation of Privileges in Multi-Agent SystemsLLM agents and tool use → Agent threats, guardrails and governance
DFlow: Enabling Verifier Information Flow in Block Diffusion Speculative DecodingEfficient LLM inference and serving → Speculative decoding
Diffuse2Seg: Diffusion Models Can Segment Anything Without SupervisionDiffusion models → Medical imaging and segmentation
OracleZoom: On-Policy Self-Distillation Inspired Reference-Constrained Recursive Image Super ResolutionKnowledge distillation → General knowledge distillation methods
Thinking with Cameras: Active Visual Reasoning via Dynamic Viewpoint Control for Surveillance Video UnderstandingLLM reasoning → Multimodal reasoning +1
Steering Under Compression: Dose-Response, Capability Cost, and Failure Asymmetry in Quantized LLMsMechanistic interpretability → Steering and interpreting LLM reasoning
One Step, One Lead: Mitigating Higher-Order Interference in Multi-Domain Reinforcement Learning via Cross-Step ControlLLM reasoning → Reinforcement learning for reasoning +1
Local and Global Stability in Performative Reinforcement LearningDeep reinforcement learning methods → RL theory and sample complexity
Causal Attribution for Agentic Decisions: Estimators, Coupling, and a Traceability SpecificationLLM agents and tool use → Software engineering coding agents
InsightChain: Optimized Chain-of-Insight Analytics for LLM-driven Data VisualizationLLM reasoning → Multimodal reasoning
On BatchNorm Forward Modes in Value-Based Reinforcement LearningDeep reinforcement learning methods → Deep RL algorithms and representations
Separating Capability from Confidence: Grounded Dual-State Calibration for GRPO-Trained Medical Vision-Language ModelsLLM reasoning → Clinical and biomedical applications +1
Visual Search Augmented Chain-of-Thought Reasoning for Attribute Value Extraction from Product VideosLLM reasoning → Multimodal reasoning
From Concentration to Differentiation and Back: Routing Effective Rank in MoE Reasoning CohortsLLM reasoning → Test-time compute and efficient thinking +1
Building Trustworthy Graph-Agentic RAG for Social Good: Architectures, Failure Propagation, and Assurance by ConstructionRetrieval-augmented generation → Agentic RAG, agents and memory +1
LANTERN: A Closed-Loop Benchmark for VLM-Based Cooperative Driving with Temporally Grounded WarningsEnd-to-end autonomous driving → Cooperative and V2X driving
Robust Conformal Consensus: Multi-Agent LLM-as-a-Judge Interval Evaluation with Conformal PredictionLLM agents and tool use → General LLM agent frameworks and benchmarks
NOVA: Normal-Side Modeling for Training-Free Zero-Shot Video Anomaly DetectionMultimodal LLMs / vision-language models → Chart understanding and anomaly detection
Robust Dynamic Expansion for Continual Learning under Backdoor Attacks via Purification and Selective RecoveryMixture-of-experts → Mixture of LoRA experts
A Ticket from Marginals to Joints: Coupled-Noise Distillation for One-Step Block Generation in Diffusion Language ModelsDiffusion models → Discrete and text diffusion
FrankenReport: Early Exiting in Long-Form Generation Using Expected Value of ComputationEfficient LLM inference and serving → Efficient reasoning and agent inference
"very likely" Means "uncertain"? How LLMs Diverge from Humans in Linguistic Uncertainty Quantification
Self-Cleaning and Captured Anyway: One Measured Primitive for Error in a Store an Agent Writes to Itself, and What a Falling Score Actually Measures
LLM-Driven Training-free Location-Attribute Synergic Fusion: A Closed-Loop Paradigm for Dual-source Encrypted POIs and LULC Mapping
FrontierMath ErdHos
AIBuildAI-2.5: Efficient Autonomous AI Model Development Through LLM-Guided Tree Search
Used, Mentioned, or Condemned? A Controlled Contrast-Set Diagnostic for the Use-Mention Distinction in Code-Mixed Hinglish Misogyny Detection
Causal Localization of the Refusal Direction in Audio Language Models
Which Part of the Context Layer Does the Work? Separating Semantic Content from Retrieval Scaffolding in Text-to-SQL Agents
Strategy Accumulation and Guided Execution for Automated LLM Fine-Tuning
CAMFT: Conflict-Aware Mergeable Fine-Tuning for Large Language Models
CALM: A Calibrated LLM Choice Network Framework for Activity-Based Traveler Simulation
Predictors and Orchestrators: Parsimonious Machine Learning within an Agentic AI Harness for Multi-Horizon Karst Aquifer Forecasting
Filling the Unseen: Scene Extrapolation via 3D Gaussian Splatting
From Masking to Merging: Rethinking SpecAugment for Efficient Audio Spectrogram Transformer
Interpretable Temporal Video Reasoning with EventGraph and EventField
A Comprehensive Review of Multimodal Facial State Analysis: Tasks, Methods, and Resources
(How) Do MLLMs Report Bistable Images Like Humans?
Halo: Improving forecast accuracy through heteroscedastic estimation
Probabilistic Focal Search: Accelerating Bounded-Suboptimal Search via Lower-Bound Advancement
Geometry Conditioning in an Embodied SLM: Training Controls and Robustness Diagnostics in a 0.8B Hybrid Model
A Subsampled Davis-Kahan Bound for Large-Scale Eigenspace Estimation
Novel Methods for Catheter and Guidewire Segmentation in X-ray Fluoroscopy under a Federated Learning Setting
MedGSSR: Generalizable Medical Image Super-Resolution 3D Reconstruction via Hierarchical Feed-forward Gaussian Splatting
Large Classification-Risk-Optional Label Acquisition
Exact Record Omission in Delta Attention: A Transport Criterion, Its Cost, and a Replay Certificate
Organization of Valence and Arousal in Vision-Language Representations of Built Environments: Insights from the EMOIS Dataset
PPIM: Pennes Physics-Informed Mamba for Heat-Source-Conditioned 3D Bioheat Simulation
Cluster-Based Dimensionality Reduction by Nonparametric Distributional Screening

All 213