Papers

New on arXiv

Tue 8 Sep 2026

451 papers

A Mechanistic Study of AI-Text Detection Neurons in Frozen BERT: Sparse Probing and Activation Patching on RAIDMechanistic interpretability → SAE features and their applications
ICDAR2026 Competition on Multimodal Reasoning over Documents in Multiple DomainsLLM reasoning → Multimodal reasoning
Grouped Value Attention: Efficient KV Caching via On-Demand Key ReconstructionEfficient LLM inference and serving → KV cache optimization
RxScribe Bench: A Multi-Axis Benchmark for Evaluating Vision-Language Models on Indian Outpatient PrescriptionsMultimodal LLMs / vision-language models → Medical vision-language models
Harness or Model? Isolating the Harness Effect in Agentic Coding with a Contamination-Controlled Private SuiteLLM agents and tool use → Software engineering coding agents
Understanding In-Context Multimodal Jailbreaks via Posterior ReweightingMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness +1
Black-Box Membership Inference via Word-Level Probability EstimationLLM safety and security → Membership inference attacks
Adaptive Diffusion Freezing: Privacy-preserving Diffusion Models Against Membership Inference AttacksDiffusion models → Robustness, privacy and adversarial aspects +1
MotionBlind: Probing the Illusion of Motion Understanding in Video-LLMsWorld models → Latent predictive world models
HLSFactory-Agent: Large-Scale Agentic HLS Dataset Construction from Academic and Open-Source ProjectsLLM agents and tool use → Software engineering coding agents
Agentic AI-enabled Semantic Commissioning of a Cognitive Digital Twin for Reconfigurable ManufacturingLLM agents and tool use → Multimodal and embodied agents
Actuator Dynamics Curricula for Narrow-Viability Tasks in Legged Robot LearningDeep reinforcement learning methods → RL for robot locomotion and manipulation
Efficient Fairness Auditing Across Guidance Scales in Text-to-Image Diffusion Models via Causal AbstractionDiffusion models → Diffusion foundations and sampling
Unthrottling the Tanh Jacobian in SAC: A Negative Result on Bang-Bang Control and MetaDriveDeep reinforcement learning methods → RL theory and sample complexity
From Fixed Keys to Readable Schemas: Small Language Models for Vehicle Agent Function CallsLLM agents and tool use → Tool learning and function calling
Code-to-Harness: Distilling Black-Box Optimizers from Self-PlaySynthetic data and self-improvement for LLMs → Self-evolving LLM agents
Low-Rank Prompt Learning for Vision-Language Models with Fixed-Token BasesMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
ContractEval: Query-Conditioned Execution Matching for Procedural Instruction ConformanceLLM agents and tool use → Software engineering coding agents
Do Agents Know When They Succeed? Calibrating Agent Confidence from Internal RepresentationsLLM agents and tool use → General LLM agent frameworks and benchmarks
DuplexJail: Safety Alignment Breaks Under Spoken Interruption in Full-Duplex ModelsLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
Valerant: An Automatic Navigable Game Map Generator via Action-Conditioned World Model ExplorationWorld models → Video-generation world models and simulators
Vision-language models know more about agriculture than they show and rubric-grounded verifications close the gapMultimodal LLMs / vision-language models → Agricultural vision-language models
An Experimental Evaluation of Multimodal Prompt Injection Attacks on Agentic AI FrameworksLLM agents and tool use → Agent threats, guardrails and governance +1
VANTAGE-Bench: Evaluating the Infrastructure AI Gap in Vision-Language ModelsMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
The Menu Is an Execution Prior: State-Path Tool Menus for Online AgentsLLM agents and tool use → Tool learning and function calling
Agentic Web Accessibility Auditing: A Criterion-Specific Framework for Translating WCAG Requirements into AssessmentsLLM agents and tool use → GUI and mobile agents
What Does MMLU Actually Measure? A Psychometric Audit of Difficulty Structure in Aggregate Benchmark ScoresLLM reasoning → Reasoning benchmarks and evaluation
Do LLMs Make More Mistakes If They Do Not Believe the Input Data?Retrieval-augmented generation → RAG robustness and adaptive retrieval
Osprey: Target-agnostic Pre-training Makes Stronger Drafters in Speculative DecodingEfficient LLM inference and serving → Speculative decoding
Video-MOPD: Multi-Teacher On-Policy Distillation for Video UnderstandingKnowledge distillation → Multimodal distillation
Steering Diffusion Priors with Sparse Observations for High-Resolution Temperature DownscalingDiffusion models → Flow matching, time series and science
In RAG We Trust? Measuring Robustness of Retrieval-Augmented Generation Under Document PoisoningRetrieval-augmented generation → Privacy and security of RAG
Distribution-Consistent Inference for Dynamic Sparse Mixture-of-ExpertsMixture-of-experts → Expert routing strategies
ReCite: Agentic Reasoning for Faithful CitationLLM agents and tool use → Web, search, and research agents
SyncWorld: Visual Calibration Enables World Models as Zero-Shot SimulatorsWorld models → Embodied and robot world models
Procedural Graphs: Self-Evolving Execution Structures for LLM AgentsLLM agents and tool use → General LLM agent frameworks and benchmarks +1
Co-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation FailsLLM agents and tool use → Software engineering coding agents
ExecCritic: Learn to Test, Test to Improve for Coding AgentsLLM agents and tool use → Software engineering coding agents
Canonical Color as a Lens into Concept Decodability in Vision Encoders and VLMsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Mask Forcing: Improving Autoregressive Video Diffusion Distillation via Dual-Noise Masking RolloutDiffusion models → Video generation
MeClear: Cooperative Game-Theoretic Attribution and Risk-Aware Memory Clearance for Long-Horizon LLM AgentsLLM agents and tool use → Agent memory systems
SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?Mechanistic interpretability → Sparse autoencoder training and evaluation
ThinkPrior: Zero-Rollout Difficulty Priors for Cold-Start Prompt Selection in RLVRLLM reasoning → Reinforcement learning for reasoning
ToolLoop: Closed-Loop Tool-Use Data Synthesis via Decomposed Generation and Dynamic Self-FeedbackLLM agents and tool use → Tool learning and function calling +1
"World Knowledge" in the Weights: Reading Concept Circuits of Vision TransformersMechanistic interpretability → Interpreting vision and multimodal models
Do Reasoning Representations Help Humans Evaluate LLM Outputs?LLM reasoning → Reasoning benchmarks and evaluation
Answer-Distribution Trajectories: A Stochastic-Dynamics View of LLM ReasoningLLM reasoning → Test-time compute and efficient thinking
Let It Go or Learn to Self-Correct: Continuous Diffusion for Constrained Discrete TasksDiffusion models → Discrete and text diffusion
Deposon: An Auditable, Conservation-Guaranteed, Game-Theoretically Tested Scattering Layer over LLM Reasoning PathsLLM reasoning → Test-time compute and efficient thinking
NERVE Attacks: Breaking AI-Powered Brain-Computer InterfacesLLM safety and security → Prompt injection and agent security
PlannerForge: LLM Agents for Scenario-Based Testing of Motion Planners in Autonomous DrivingEnd-to-end autonomous driving → Large language models for driving +1
SkillAdam: Stable and Efficient Skill Evolution for AgentsLLM agents and tool use → Agent skill learning and evolution
Experience Funnel: A State-Policy Alternating Loop for Self-Evolving AgentsLLM agents and tool use → General LLM agent frameworks and benchmarks +1
The BatchNorm Illusion: Diagnosing Normalization Artifacts in Machine Unlearning EvaluationLLM safety and security → Machine unlearning
Q2D-Web: A Large-Scale Benchmark for Retrieval in Agentic RAG SystemsRetrieval-augmented generation → Agentic RAG, agents and memory +1
Medical AI Encodes a "Feeling of Error": Verifying Cancer Segmentation via Internal ConceptsMechanistic interpretability → Interpreting biomedical and science models
SeGDeP: Semantic- and Geometric-Aware Decoupled Prompts for Reasoning SegmentationLLM reasoning → Multimodal reasoning
Earth System World Model for What-If Simulations: A Case Study for Terrestrial EcosystemsWorld models → Latent predictive world models
CAST: Alternating State-Value Targets and Expanded Policy Gradients for Model-Based Reinforcement LearningDeep reinforcement learning methods → RL for robot locomotion and manipulation
Closing the Consistency Gap: Self-Evolving Agents That Learn to Stay on CourseLLM agents and tool use → General LLM agent frameworks and benchmarks +1
Eliciting Weak-to-Strong Generalization with On-Policy Reverse DistillationKnowledge distillation → On-policy and self-distillation for LLMs
Hi-FLoop: Hierarchical State-Feedback Loops for Multi-Timescale World ModelingWorld models → World models for autonomous driving
MemSentry: A Framework for Detecting Persistent Memory Poisoning in Agentic AILLM agents and tool use → Agent threats, guardrails and governance
Global Divergence, Local Convergence: Representation Geometry in SSMs and TransformersState-space sequence models (S4, Mamba) → Core SSM architectures for language
Hyperparameter Scaling Laws Across MoE SparsityMixture-of-experts → MoE theory and gating functions
HDA-MoE: Hybrid Parallelism and Dynamic, Adaptive Scheduling for Mixture-of-Experts with 3D Near-Memory ProcessingMixture-of-experts → Efficient MoE training and inference systems
Learning to build covering structures with continuous adjustmentsDeep reinforcement learning methods → RL for robot locomotion and manipulation
MoEMB: Scaling Universal Multimodal Embeddings with Efficient Mixture-of-Experts ModelsMixture-of-experts → Multimodal MoE models
Difficulty-Adaptive Tree-Structured Policy Optimization for Expanding Reasoning Coverage in RLVRLLM reasoning → Reinforcement learning for reasoning
SUN: Reaching for Novelty in Reinforcement LearningDeep reinforcement learning methods → Deep RL algorithms and representations
Suan: Rectifying Direct Preference Safety Alignment in Large Language ModelsLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
CLAMP: Constrained Decoding for Vision-Language Embodied PlanningMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Graph-Based Personalized Memory for LLM Agents: Representation, Evolution, Retrieval, and EvaluationLLM agents and tool use → Agent memory systems
A Three-Tier Persona Vector for Controllable User Simulation in Agentic EvaluationLLM agents and tool use → Social simulation and multi-agent debate
The Unreliable Progress Bar: Can LLM Agents Reliably Report Task Progress Throughout Execution?LLM agents and tool use → Software engineering coding agents
Estimating Semantic Ambiguity via Gaussian Context Distributions for VLM-Driven Traversability AnalysisMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Which Forms of Caregiver Feedback Support Grammar Learning? A Reinforcement-Learning Study of Child-Like Language ModelsLLM alignment / RLHF / preference optimisation → Reward modeling and RLHF
AgentGrad: Intervention-guided Prompt Optimization for Multi Agent SystemsLLM agents and tool use → General LLM agent frameworks and benchmarks
BIO-MEMART: Biometric-Aware KV Cache Memory for Multi-User LLM AgentsLLM agents and tool use → Agent memory systems +1
Personalizing LLM Agent Memory Using BiometricsLLM agents and tool use → Agent memory systems
CreaMem: A Scene-Aware Memory Architecture for Personalized AgentsLLM agents and tool use → Agent memory systems
STSG-VQA: Evidence-Grounded Temporal Question Answering from Surgical Spatio-Temporal Scene GraphsLLM reasoning → Clinical and biomedical applications +1
Concept-Level Risk and Calibration for Governance in Diffusion Foundation ModelsDiffusion models → Robustness, privacy and adversarial aspects
SignRefine: Adapting Foundational Video Models for Sign Language GenerationDiffusion models → Sign language production
An Evidence Model for Agentic Processes: Evidence Claims, Trust Assumptions, and Policy AssessmentLLM agents and tool use → Agent threats, guardrails and governance
Do Reviewers Still Reward Lexical Complexity? A Frozen-Rater Study of Preference Drift in 124K ICLR ReviewsLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
Beyond Agent Harnesses: Cross-Substrate Authority for Multi-Agent SystemsLLM agents and tool use → Agent threats, guardrails and governance
SRPO: Setwise Relative Policy Optimization for Multi-Agent LLMsLLM agents and tool use → Reinforcement learning for LLM agents
GSComplete: Gaussian Splat Completion with 2D Diffusion PriorsDiffusion models → 3D, point cloud and scene generation
Safe Task Planning with Long-Term Graph Memory for Embodied AgentsMultimodal LLMs / vision-language models → Embodied and 3D vision-language models +1
A Two-Stage, Model-Based Reinforcement Learning Approach for Active Flow Control of Bluff Body WakesDeep reinforcement learning methods → RL with diffusion and flow models
Compositional Multilingual and Behavioral Attribute SteeringLLM safety and security → Safety alignment, refusal and guardrails +1
FastE: Readout-Triggered Token Compression for LLM Embedding InferenceEfficient LLM inference and serving → KV cache optimization
Environments as Scaffold: Enriching Feedback to Bootstrap Self-Evolving Agents in Long-Horizon TasksLLM agents and tool use → Reinforcement learning for LLM agents +1
Towards Embodied Air-Ground Cooperative Object Search: Benchmark, Dataset and Agentic MethodLLM agents and tool use → Multimodal and embodied agents
From Coordinates to Candidate Regions: Temporal Change Localization via Region Selection in Remote Sensing Multimodal LLMsMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Structural Jailbreaks Generalize but Do Not Compound: A cross-provider and multilingual study of Involuntary In-Context LearningLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
Authority Is Not a String: A Capability-Scoped Harness for Prompt-Injection-Resistant Coding AgentsLLM agents and tool use → Agent threats, guardrails and governance
To Adapt or Not to Adapt? Selective Adaptation for Vision-Language ModelsMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
ReMoMask-2: Latent Retrieval-Augmented Masked Motion GenerationRetrieval-augmented generation → Multimodal RAG

All 451