Papers

New on arXiv

Sun 3 May 2026

229 papers

Securing Multimodal AI through Internal Information DecompositionMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness
Decoupled Attention Fusion: Accelerating RAG with Efficient KV Cache ReuseRetrieval-augmented generation → Generation and context integration +1
HALO: Hybrid Adaptive Latent Reasoning for Language ModelsLLM reasoning → Test-time compute and efficient thinking
Benchmarking KV-Cache Optimizations across Task Quality and System Performance for Long-Context ServingEfficient LLM inference and serving → KV cache optimization
Human-Less LLM Serving: Quantifying the Human Tax on ThroughputEfficient LLM inference and serving → Serving systems and request scheduling
Preregistration for Experiments with AI AgentsLLM agents and tool use → Social simulation and multi-agent debate
Accelerating Long-Tail Generation in Synchronous RLHF Training via Adaptive Tensor ParallelismLLM alignment / RLHF / preference optimisation → Reward modeling and RLHF
Feature Rivalry in Sparse Autoencoder Representations: A Mechanistic Study of Uncertainty-Driven Feature Competition in LLMsMechanistic interpretability → SAE features and their applications
Self-Captioning Multimodal Interaction Tuning: Amplifying Exploitable Redundancies for Robust Vision Language ModelsMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
Beyond Semantic Similarity: Rethinking Retrieval for Agentic Search via Direct Corpus InteractionLLM agents and tool use → Web, search, and research agents
DexSim2Real: Foundation Model-Guided Sim-to-Real Transfer for Generalizable Dexterous ManipulationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Structured Diffusion Bridges: Inductive Bias for Denoising Diffusion BridgesDiffusion models → Image generation, text-to-image and editing
Multilingual Safety Alignment via Self-DistillationLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +2
AutoRAGTuner: A Declarative Framework for Automatic Optimization of RAG PipelinesRetrieval-augmented generation → RAG pipeline design and evaluation
Reward Hacking Benchmark: Measuring Exploits in LLM Agents with Tool UseLLM agents and tool use → Reinforcement learning for LLM agents +1
ISAAC: Auditing Causal Reasoning in Deep Models for Drug-Target InteractionLLM reasoning → Clinical and biomedical applications
MoE-Prefill: Zero Redundancy Overheads in MoE Prefill ServingMixture-of-experts → Efficient MoE training and inference systems
NORA: A Harness-Engineered Autonomous Research Agent for End-to-End Spatial Data ScienceLLM agents and tool use → Web, search, and research agents
Enhanced LLM Reasoning by Optimizing Reward Functions with Search-Driven Reinforcement LearningLLM reasoning → Reinforcement learning for reasoning
VILAS: A VLA-Integrated Low-cost Architecture with Soft Grasping for Robotic ManipulationVision-language-action models → Tactile, force-aware and humanoid VLA
Conventional Commit Classification using Large Language Models and Prompt EngineeringLLM reasoning → Reasoning for agents, planning and code
Enhancing Judgment Document Generation via Agentic Legal Information Collection and Rubric-Guided OptimizationLLM reasoning → Reasoning for agents, planning and code +1
12 Angry AI Agents: Evaluating Multi-Agent LLM Decision-Making Through Cinematic Jury DeliberationLLM agents and tool use → Social simulation and multi-agent debate
Stability of Control Lyapunov Function Guided Reinforcement LearningDeep reinforcement learning methods → RL for robot locomotion and manipulation
Trojan Hippo: Weaponizing Agent Memory for Data ExfiltrationLLM agents and tool use → Agent threats, guardrails and governance
AdamO: A Collapse-Suppressed Optimizer for Offline RLOffline RL and imitation learning from data → Offline RL algorithms
Multi-User Dueling Bandits: A Fair Approach using Nash Social WelfareLLM alignment / RLHF / preference optimisation → Reward modeling and RLHF
Flexi-LoRA with Input-Adaptive Ranks: Efficient Finetuning for Speech and Reasoning TasksLLM reasoning → Test-time compute and efficient thinking
Moira: Language-driven Hierarchical Reinforcement Learning for Pair TradingDeep reinforcement learning methods → Adversarial robustness and RL in finance
TRAP: Tail-aware Ranking Attack for World-Model PlanningWorld models → World model security and GUI agents +1
Phone2Act: A Low-Cost, Hardware-Agnostic Teleoperation System for Scalable VLA Data CollectionVision-language-action models → Tactile, force-aware and humanoid VLA
StressEval: Failure-Driven Dynamic Benchmarking for Knowledge-Intensive Reasoning in Large Language ModelsLLM reasoning → Test-time compute and efficient thinking
ViM-Q: Scalable Algorithm-Hardware Co-Design for Vision Mamba Model Inference on FPGAState-space sequence models (S4, Mamba) → Vision Mamba backbones
Exploring Data-Free LoRA Transferability for Video Diffusion ModelsDiffusion models → Video generation
CADFS: A Big CAD Program Dataset and Framework for Computer-Aided Design with Large Language ModelsMultimodal LLMs / vision-language models → Chart understanding and anomaly detection
A Language for Describing Agentic LLM ContextsLLM agents and tool use → General LLM agent frameworks and benchmarks
RefusalGuard: Geometry-Preserving Fine-Tuning for Safety in LLMsLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks
SurgCheck: Do Vision-Language Models Really Look at Images in Surgical VQA?Multimodal LLMs / vision-language models → Medical vision-language models
Stochastic Sparse Attention for Memory-Bound InferenceEfficient LLM inference and serving → KV cache optimization
Disentangling Intent from Role: Adversarial Self-Play for Persona-Invariant Safety AlignmentLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +2
Divide and Conquer: Decoupled Representation Alignment for Multimodal World ModelsWorld models → Video-generation world models and simulators
QASecClaw: A Multi-Agent LLM Approach for False Positive Reduction in Static Application Security TestingLLM agents and tool use → Agent threats, guardrails and governance
Chart-FR1: Visual Focus-Driven Fine-Grained Reasoning on Dense ChartsLLM reasoning → Multimodal reasoning +1
Maistros: A Greek Large Language Model Adapted Through Knowledge Distillation From Large Reasoning ModelsLLM reasoning → Reasoning benchmarks and evaluation +1
QHyer: Q-conditioned Hybrid Attention-mamba Transformer for Offline Goal-conditioned RLState-space sequence models (S4, Mamba) → Video, motion and generative Mamba models
Decouple and Cache: KV Cache Construction for Streaming Video UnderstandingEfficient LLM inference and serving → Visual token pruning in multimodal models
Spatiotemporal Hidden-State Dynamics as a Signature of Internal Reasoning in Large Language ModelsLLM reasoning → Reasoning benchmarks and evaluation
NeuroState-Bench: A Human-Calibrated Benchmark for Commitment Integrity in LLM Agent ProfilesLLM agents and tool use → Software engineering coding agents
RMGAP: Benchmarking the Generalization of Reward Models across Diverse PreferencesLLM alignment / RLHF / preference optimisation → Reward modeling and RLHF
GeoSAE: Geometric Prior-Guided Layer-Wise Sparse Autoencoder Annotation of Brain MRI Foundation ModelsMechanistic interpretability → Interpreting biomedical and science models
Referring Multiple Regions with Large Multimodal Models via Contextual Latent SteeringMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
Selector-Guided Autonomous Curriculum for One-Shot Reinforcement Learning from Verifiable RewardsLLM reasoning → Reinforcement learning for reasoning
Skipping the Zeros in Diffusion Models for Sparse Data GenerationDiffusion models → Diffusion foundations and sampling
Embody4D: A Generalist Data Engine for Embodied 4D World ModelingWorld models → Embodied and robot world models +1
Needle-in-RAG: Prompt-Conditioned Character-Level Traceback of Poisoned Spans in Retrieved EvidenceRetrieval-augmented generation → Privacy and security of RAG
MedScribe: Clinically Grounded CT Reporting through Agentic WorkflowsMultimodal LLMs / vision-language models → Medical vision-language models
Adversarial Imitation Learning with General Function Approximation: Theoretical Analysis and Practical AlgorithmsOffline RL and imitation learning from data → Imitation learning methods
Anticipation-VLA: Solving Long-Horizon Embodied Tasks via Anticipation-based Subgoal GenerationVision-language-action models → Generalist VLA models and benchmarks
Mitigating Multimodal LLMs Hallucinations via Relevance Propagation at Inference TimeMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
TrajShield: Trajectory-Level Safety Mediation for Defending Text-to-Video Models Against Jailbreak AttacksLLM safety and security → Vision-language and text-to-image safety
Catching the Infection Before It Spreads: Foresight-Guided Defense in Multi-Agent SystemsLLM safety and security → Prompt injection and agent security
Talk is Cheap, Communication is Hard: Dynamic Grounding Failures and Repair in Multi-Agent NegotiationLLM agents and tool use → Social simulation and multi-agent debate
Only Say What You Know: Calibration-Aware Generation for Long-Form FactualityLLM reasoning → Test-time compute and efficient thinking
Architectural Obsolescence of Unhardened Agentic-AI RuntimesLLM agents and tool use → Agent threats, guardrails and governance
AgenticVM: Agentic AI for Adaptive Software Vulnerability ManagementLLM agents and tool use → Agent threats, guardrails and governance
Less is More: Geometric Unlearning for LLMs with Minimal Data DisclosureLLM safety and security → Machine unlearning
GEASS: Gated Evidence-Adaptive Selective Caption Trust for Vision-Language ModelsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
SplitZip: Ultra Fast Lossless KV Compression for Disaggregated LLM ServingEfficient LLM inference and serving → KV cache optimization
The Reasoning Trap: An Information-Theoretic Bound on Closed-System Multi-Step LLM ReasoningLLM reasoning → Reasoning for agents, planning and code
TrajRAG: Retrieving Geometric-Semantic Experience for Zero-Shot Object NavigationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Probe-Geometry Alignment: Erasing the Cross-Sequence Memorization Signature Below ChanceLLM safety and security → Machine unlearning
Latent State Design for World Models under Sufficiency ConstraintsWorld models → World models for language agents
MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM SafetyLLM safety and security → Jailbreaking
CP-SynC: Multi-Agent Zero-Shot Constraint Modeling in MiniZinc with Synthesized CheckersLLM agents and tool use → General LLM agent frameworks and benchmarks
A Hybrid Retrieval and Reranking Framework for Evidence-Grounded Retrieval-Augmented GenerationRetrieval-augmented generation → Medical and biomedical RAG
Towards Efficient and Expressive Offline RL via Flow-Anchored Noise-conditioned Q-LearningOffline RL and imitation learning from data → Diffusion and flow policies for offline RL
Video Active Perception: Effective Inference-Time Long-Form Video Understanding with Vision-Language ModelsMultimodal LLMs / vision-language models → Video understanding with MLLMs
TRIMMER: A New Paradigm for Video Summarization through Self-Supervised Reinforcement LearningDeep reinforcement learning methods → RL training of language and agent models
Act2See: Emergent Active Visual Perception for Video ReasoningLLM reasoning → Multimodal reasoning +1
SteeringDiffusion: A Bottlenecked Activation Control Interface for Diffusion ModelsDiffusion models → Diffusion foundations and sampling
Auxiliary uncertainty signals for LLM-assisted systematic review screening: a benchmark across eight Cohen drug-class reviews
From Frame-Level Recognition to Event-Level Confirmation: Repair Traces and Runtime Failure Analysis of Public-Space Gesture Interaction
FixItFlow: Automated Troubleshooting Guide Generation from Cloud Incidents
How Personas Can Influence Agents to Play Split or Steal
Prompt Framing Distorts Count Based Evaluation of LLM Error Detection: Evidence from Numeric Anchoring
The Environmental Cost of LLMs in AIED: Reporting and Practices
Progressive in Principle, Centrist in Practice: LLM Political Bias Is Instrument-Dependent
Information as Maximum-Caliber Deviation: A bridge between Integrated Information Theory and the Free Energy Principle
A PyTorch Library of Turing-Complete Neural Networks
VT-Bench: A Unified Benchmark for Visual-Tabular Multi-Modal Learning
PPO-Based Dynamic Positioning of HAPS-BS in Wind-Disturbed Stratospheric Maritime Networks
HERCULES: Hardware-Efficient, Robust, Continual Learning Neural Architecture Search
Meta-LegNet: A Transferable and Interpretable Framework for Surface Adsorption Prediction via Self-Defined Adsorption-Environment Learning
PHBench: A Benchmark for Predicting Startup Series A Funding from Product Hunt Launch Signals
Decompose to Understand, Fuse to Detect: Frequency-Decoupled Anomaly Detection for Encrypted Network Traffic
Finite-Size Gradient Transport in Large Language Model Pretraining: From Cascade Size to Intensive Transport Efficiency
Joint Energy Management and Coordinated AIGC Workload Scheduling for Distributed Data Centers: A Diffusion-Aided Reward Shaping Approach
Analytic Bridge Diffusions for Controlled Path Generation
Calibration of the underlying surface parameters for urban flood using latent variables and adjoint equation
Bridging the Gap Between Average and Discounted TD Learning

All 229