Papers

New on arXiv

Sat 9 May 2026

398 papers

JUMP: Single-Pass Membership Inference on Fine-Tuned Diffusion Language ModelsDiffusion models → Discrete and text diffusion +1
The Interference Gap: Comparing Retrieval Bounds in Human Memory and RAG SystemsRetrieval-augmented generation → RAG robustness and adaptive retrieval
Where Larger Models Excel: The Primacy of Constraint-Guided ReasoningLLM reasoning → Mathematical reasoning
Mechanistic Analysis of Alignment Algorithms in Language ModelsLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Methods for Formal Verification of Agent Skills: Three Layers Toward a Mechanically Checkable Capability-Containment ProofLLM agents and tool use → Agent skill learning and evolution
When Is Rank-1 Steering Cheap? Geometry, Granularity, and Budgeted SearchMechanistic interpretability → Activation steering methods
How Many Visual Tokens Do Multimodal Language Models Need? Scaling Visual Token Pruning with F^3AMultimodal LLMs / vision-language models → Visual token pruning and efficiency +1
SkillGen: Verified Inference-Time Agent Skill SynthesisLLM agents and tool use → Agent skill learning and evolution
Few-Shot Truly Benign DPO Attack for Jailbreaking LLMsLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
ECHO: Continuous Hierarchical Memory for Vision-Language-Action ModelsVision-language-action models → Reasoning and chain-of-thought in VLA
Test-Time Personalization: A Diagnostic Framework and Probabilistic Fix for Scaling FailuresLLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment
Skill Drift Is Contract Violation: Proactive Maintenance for LLM Agent Skill LibrariesLLM agents and tool use → Agent skill learning and evolution
TMPO: Trajectory Matching Policy Optimization for Diverse and Efficient Diffusion AlignmentDiffusion models → Diffusion and flow policies for robotics +1
xi-DPO: Direct Preference Optimization via Ratio Reward MarginLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
LEAP: Unlocking dLLM Parallelism via Lookahead Early-Convergence Token DetectionDiffusion models → Discrete and text diffusion +1
PASA: A Principled Embedding-Space Watermarking Approach for LLM-Generated Text under Semantic-Invariant AttacksLLM safety and security → Watermarking generated content and models
The Art of the Jailbreak: Formulating Jailbreak Attacks for LLM Security Beyond Binary ScoringLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
SMIXAE: Towards Unsupervised Manifold Discovery in Language ModelsMechanistic interpretability → Sparse autoencoder training and evaluation
Flame3D: Zero-shot Compositional Reasoning of 3D Scenes with Agentic Language ModelsLLM reasoning → Spatial and embodied reasoning +1
Fast Rates for Offline Contextual Bandits with Forward-KL Regularization under Single-Policy ConcentrabilityDeep reinforcement learning methods → RL theory and sample complexity
LBI: Parallel Scan Backpropagation via Latent Bounded InterfacesState-space sequence models (S4, Mamba) → Core SSM architectures for language
Removing the Watermark Is Not Enough: Forensic Stealth in Generative-AI Watermark RemovalLLM safety and security → Watermarking generated content and models
DARE: Difficulty-Adaptive Reinforcement Learning with Co-Evolved Difficulty EstimationLLM reasoning → Reinforcement learning for reasoning +1
Agentic MIP Research: Accelerated Constraint Handler GenerationLLM agents and tool use → General LLM agent frameworks and benchmarks
Learning When to Stop: Selective Imitation Learning Under Arbitrary Dynamics ShiftOffline RL and imitation learning from data → Imitation learning methods
CIVeX: Causal Intervention Verification for Language AgentsLLM agents and tool use → Medical and scientific domain agents
Sparse Layers are Critical to Scaling Looped Language ModelsMixture-of-experts → MoE large language models +1
Data-Driven Inverse Reinforcement Learning of Linear Systems with Model Uncertainty: A Convex Optimization ViewOffline RL and imitation learning from data → Inverse reinforcement learning +1
FORTIS: Benchmarking Over-Privilege in Agent SkillsLLM agents and tool use → Agent skill learning and evolution
Do LLMs Experience an Internal Polylogue? Investigating Reasoning through the Lens of PersonasLLM reasoning → Reasoning benchmarks and evaluation
Revisiting Mixture Policies in Entropy-Regularized Actor-CriticDeep reinforcement learning methods → RL theory and sample complexity
Meow-Omni 1: A Multimodal Large Language Model for Feline EthologyMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
BoostAPR: Boosting Automated Program Repair via Execution-Grounded Reinforcement Learning with Dual Reward ModelsLLM alignment / RLHF / preference optimisation → Process reward models for reasoning
MCP-Cosmos: World Model-Augmented Agents for Complex Task Execution in MCP EnvironmentsWorld models → World models for language agents +1
Data-driven Circuit Discovery for Interpretability of Language ModelsMechanistic interpretability → Circuits and mechanistic analysis
Transfer Learning of Multiobjective Indirect Low-Thrust Trajectories Using Diffusion Models and Markov Chain Monte CarloDiffusion models → Diffusion and flow policies for robotics
A Communication-Theoretic Framework for LLM Agents: Cost-Aware Adaptive ReliabilityLLM reasoning → Reasoning for agents, planning and code +1
When (and How) to Trust the Expert: Diagnosing Query-Time Expert-Guided Reinforcement LearningDeep reinforcement learning methods → Deep RL algorithms and representations
Token Economics for LLM Agents: A Dual-View Study from Computing and EconomicsLLM agents and tool use → Social simulation and multi-agent debate
GRC: Unifying Reasoning-Driven Generation, Retrieval and CompressionLLM reasoning → Test-time compute and efficient thinking +1
CauSim: Scaling Causal Reasoning with Increasingly Complex Causal SimulatorsLLM reasoning → Causal reasoning in LLMs
Robust Multi-Agent LLMs under Byzantine FaultsLLM agents and tool use → General LLM agent frameworks and benchmarks
Probability-Flow Distillation: Distribution Matching in Parameter SpaceDiffusion models → Diffusion foundations and sampling
Single-Configuration Attack Success Rate Is Not Enough: Jailbreak Evaluations Should Report Distributional Attack SuccessLLM safety and security → Jailbreaking
Dependency-Aware Discrete Diffusion for Scene Graph GenerationDiffusion models → Image generation, text-to-image and editing
Soohak: A Mathematician-Curated Benchmark for Evaluating Research-level Math Capabilities of LLMsLLM reasoning → Mathematical reasoning
Language-Conditioned Visual Grounding with CLIP MultilingualMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Containment Verification: AI Safety Guarantees Independent of AlignmentLLM agents and tool use → Agent threats, guardrails and governance
Evaluating Pragmatic Reasoning in Large Language Models: Evidence from Scalar DiversityLLM reasoning → Reasoning benchmarks and evaluation
ShadowMerge: A Novel Poisoning Attack on Graph-Based Agent Memory via Relation-Channel ConflictsLLM reasoning → Reasoning for agents, planning and code +1
GAMBIT: A Three-Mode Benchmark for Adversarial Robustness in Multi-Agent LLM CollectivesLLM agents and tool use → Agent threats, guardrails and governance
Evolving Ensemble of AgentsLLM agents and tool use → Software engineering coding agents
Re2Math: Benchmarking Theorem Retrieval in Research-Level MathematicsLLM reasoning → Mathematical reasoning
Relative Kinetic Utility for Reasoning-Aware Structural Pruning in Large Language ModelsLLM reasoning → Test-time compute and efficient thinking
Towards Backdoor-Based Ownership Verification for Vision-Language-Action ModelsVision-language-action models → Safety and adversarial attacks +1
LLaVA-UHD v4: What Makes Efficient Visual Encoding in MLLMs?Multimodal LLMs / vision-language models → Visual token pruning and efficiency +1
Learning to Explore: Scaling Agentic Reasoning via Exploration-Aware Policy OptimizationLLM reasoning → Reasoning for agents, planning and code +1
Tracking the Truth: Object-Centric Spatio-Temporal Monitoring for Video Large Language ModelsMultimodal LLMs / vision-language models → Video understanding with MLLMs
Can MLLMs Reason About Visual Persuasion? Evaluating the Efficacy and Faithfulness of ReasoningLLM reasoning → Multimodal reasoning +1
Trustworthy AI: Ensuring Reliability and Accountability from Models to AgentsLLM agents and tool use → Agent threats, guardrails and governance +1
Agentic AI Scientists Are Not Built For Autonomous Scientific DiscoveryLLM agents and tool use → Web, search, and research agents
A Single Deep Preference-Conditioned Policy for Learning Pareto Coverage SetsDeep reinforcement learning methods → RL for games and combinatorial optimization
MDGYM: Benchmarking AI Agents on Molecular SimulationsLLM agents and tool use → Web, search, and research agents
Self-ReSET: Learning to Self-Recover from Unsafe Reasoning TrajectoriesLLM reasoning → Reasoning for agents, planning and code
From Mechanistic to Compositional InterpretabilityMechanistic interpretability → Circuits and mechanistic analysis
Internalizing Safety Understanding in Large Reasoning Models via VerificationLLM reasoning → Reasoning for agents, planning and code +1
Non-Monotonic Latency in Apple MPS Decoding: KV Cache Interactions and Execution RegimesEfficient LLM inference and serving → KV cache optimization
OPT-BENCH: Evaluating the Iterative Self-Optimization of LLM Agents in Large-Scale Search SpacesLLM agents and tool use → General LLM agent frameworks and benchmarks
DAPE: Dynamic Non-uniform Alignment and Progressive Detail Enhancement Techniques for Improving the Performance of Efficient Visual Language ModelsMultimodal LLMs / vision-language models → MLLM training and visual instruction tuning
LLM-Agnostic Semantic Representation AttackLLM safety and security → Jailbreaking
Bilinear autoencoders find interpretable manifoldsMechanistic interpretability → Sparse autoencoder training and evaluation
DocScope: Benchmarking Verifiable Reasoning for Trustworthy Long-Document UnderstandingLLM reasoning → Multimodal reasoning +1
Discrete Flow Matching: Convergence Guarantees Under Minimal AssumptionsDiffusion models → Flow matching, time series and science
Preserving Foundational Capabilities in Flow-Matching VLAs through Conservative SFTVision-language-action models → Efficient VLA inference and adaptation +1
Why Do Aligned LLMs Remain Jailbreakable: Refusal-Escape Directions, Operator-Level Sources, and Safety-Utility Trade-offLLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks +1
OTora: A Unified Red Teaming Framework for Reasoning-Level Denial-of-Service in LLM AgentsLLM reasoning → Reasoning for agents, planning and code +2
Semantic Alignment in Hyperbolic Space for Open-Vocabulary Semantic SegmentationMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
CoDistill-GRPO: A Co-Distillation Recipe for Efficient Group Relative Policy OptimizationLLM reasoning → Reinforcement learning for reasoning
BubbleSpec: Turning Long-Tail Bubbles into Speculative Rollout Drafts for Synchronous Reinforcement LearningDeep reinforcement learning methods → RL training of language and agent models
Architecture, Not Scale: Circuit Localization in Large Language ModelsMechanistic interpretability → Circuits and mechanistic analysis
XPERT: Expert Knowledge Transfer for Effective Training of Language ModelsMixture-of-experts → MoE large language models
Illusion-Aware Visual Preprocessing and Anti-Illusion Prompting for Classic Illusion Understanding in Vision-Language ModelsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
ReST-KV: Robust KV Cache Eviction with Layer-wise Output Reconstruction and Spatial-Temporal SmoothingEfficient LLM inference and serving → KV cache optimization
Generating Leakage-Free Benchmarks for Robust RAG EvaluationRetrieval-augmented generation → RAG pipeline design and evaluation
The Grounding Gap: How LLMs Anchor the Meaning of Abstract Concepts Differently from HumansMechanistic interpretability → Knowledge and representations inside LLMs
SynerDiff: Synergetic Continuous Batching for Fast and Parallel Diffusion Model InferenceDiffusion models → Diffusion foundations and sampling +1
VECTOR-Drive: Tightly Coupled Vision-Language and Trajectory Expert Routing for End-to-End Autonomous DrivingVision-language-action models → VLA for autonomous driving +2
When Agents Overtrust Environmental Evidence: An Extensible Agentic Framework for Benchmarking Evidence-Grounding Defects in LLM AgentsLLM agents and tool use → Agent threats, guardrails and governance
How You Begin is How You Reason: Driving Exploration in RLVR via Prefix-Tuned PriorsLLM reasoning → Reinforcement learning for reasoning +1
Mirror, Mirror on the Wall: Can VLM Agents Tell Who They Are at All?LLM agents and tool use → Multimodal and embodied agents
AgentSlimming: Towards Efficient and Cost-Aware Multi-Agent SystemsLLM agents and tool use → General LLM agent frameworks and benchmarks
SimReg: Achieving Higher Performance in the Pretraining via Embedding Similarity RegularizationMixture-of-experts → MoE large language models
Constraint-Aware Diffusion Priors for High-Fidelity and Versatile Quadruped LocomotionDiffusion models → Diffusion and flow policies for robotics +2
CoLVR: Enhancing Exploratory Latent Visual Reasoning via Contrastive OptimizationLLM reasoning → Multimodal reasoning +1
PPU-Bench:Real World Benchmark for Personalized Partial Unlearning in Vision Language ModelsMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness +1
ElasticFlow: One-Step Physics-Consistent Policy with Elastic Time Horizons for Language-Guided ManipulationDiffusion models → Diffusion and flow policies for robotics
Lost in Volume: The CT-SpatialVQA Benchmark for Evaluating Semantic-Spatial Understanding of 3D Medical Vision-Language ModelsLLM reasoning → Clinical and biomedical applications +1
Not All Turns Matter: Credit Assignment for Multi-Turn JailbreakingLLM safety and security → Jailbreaking
Reasoning Compression with Mixed-Policy DistillationLLM reasoning → Test-time compute and efficient thinking +1
ProcVLM: Learning Procedure-Grounded Progress Rewards for Robotic ManipulationLLM alignment / RLHF / preference optimisation → Reward modeling and RLHF

All 398