Papers

New on arXiv

Fri 16 May 2025

505 papers

SpecMemo: Speculative Decoding is in Your PocketEfficient LLM inference and serving → Speculative decoding
SpecBranch: Speculative Decoding via Hybrid Drafting and Rollback-Aware Branch ParallelismEfficient LLM inference and serving → Speculative decoding
Social preferences with unstable interactive reasoning: Large language models in economic trust gamesLLM reasoning → Reasoning for agents, planning and code
SpecEdge: Scalable Edge-Assisted Serving Framework for Interactive LLMsEfficient LLM inference and serving → Serving systems and request scheduling
Time-R1: Towards Comprehensive Temporal Reasoning in LLMsLLM reasoning → Knowledge- and tool-augmented reasoning
Open Set Domain Adaptation with Vision-language models via Gradient-aware SeparationMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
EcoSafeRAG: Efficient Security through Context Analysis in Retrieval-Augmented GenerationRetrieval-augmented generation → Privacy and security of RAG
An agentic system with reinforcement-learned subsystem improvements for parsing form-like documentsLLM agents and tool use → Web, search, and research agents
SPIEDiff: robust learning of long-time macroscopic dynamics from short-time particle simulations with quantified epistemic uncertaintyDiffusion models → Diffusion foundations and sampling
Noise Injection Systemically Degrades Large Language Model Safety GuardrailsLLM safety and security → Safety alignment, refusal and guardrails
IRLBench: A Multi-modal, Culturally Grounded, Parallel Irish-English Benchmark for Open-Ended LLM Reasoning EvaluationLLM reasoning → Reasoning benchmarks and evaluation
Generalizable Vision-Language Few-Shot Adaptation with Predictive Prompts and Negative LearningMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
Feature Hedging: Correlated Features Break Narrow Sparse AutoencodersMechanistic interpretability → Sparse autoencoder training and evaluation
X-Edit: Detecting and Localizing Edits in Images Altered by Text-Guided Diffusion ModelsDiffusion models → Image generation, text-to-image and editing
Missing Data Imputation by Reducing Mutual Information with Rectified FlowsDiffusion models → Flow matching, time series and science
MTRE: Multi-Token Reliability Estimation for Hallucination Detection in VLMsMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
TokUR: Token-Level Uncertainty Estimation for Large Language Model ReasoningLLM reasoning → Test-time compute and efficient thinking
MedCaseReasoning: Evaluating and learning diagnostic reasoning from clinical case reportsLLM reasoning → Clinical and biomedical applications
Rethinking Optimal Verification Granularity for Compute-Efficient Test-Time ScalingLLM reasoning → Test-time compute and efficient thinking
Zero-Shot Visual Generalization in Robot ManipulationOffline RL and imitation learning from data → Robot manipulation imitation learning
REMOR: Automated Peer Review Generation with LLM Reasoning and Multi-Objective Reinforcement LearningLLM reasoning → Knowledge- and tool-augmented reasoning
WebInject: Prompt Injection Attack to Web AgentsLLM agents and tool use → Agent threats, guardrails and governance +1
Bi-Level Policy Optimization with Nystr"om HypergradientsDeep reinforcement learning methods → RL theory and sample complexity
Unveiling the Black Box: A Multi-Layer Framework for Explaining Reinforcement Learning-Based Cyber AgentsDeep reinforcement learning methods → Adversarial robustness and RL in finance
Attend to Not Attended: Structure-then-Detail Token Merging for Post-training DiT AccelerationDiffusion models → Diffusion foundations and sampling +1
Ambiguity in LLMs is a concept missing problemLLM agents and tool use → General LLM agent frameworks and benchmarks
Terminators: Terms of Service Parsing and Auditing AgentsLLM agents and tool use → Social simulation and multi-agent debate
Learning from Less: Guiding Deep Reinforcement Learning with Differentiable Symbolic PlanningDeep reinforcement learning methods → Interpretable and logic-guided RL
MIRACL-VISION: A Large, multilingual, visual document retrieval benchmarkRetrieval-augmented generation → Multimodal RAG
Can an Easy-to-Hard Curriculum Make Reasoning Emerge in Small Language Models? Evidence from a Four-Stage Curriculum on GPT-2LLM reasoning → Prompting and chain-of-thought reasoning
PeerGuard: Defending Multi-Agent Systems Against Backdoor Attacks Through Mutual ReasoningLLM reasoning → Reasoning for agents, planning and code +2
The Gaussian-Multinoulli Restricted Boltzmann Machine: A Potts Model Extension of the GRBMLLM reasoning → Test-time compute and efficient thinking
Critique-Guided Distillation for Robust Reasoning via RefinementLLM reasoning → Test-time compute and efficient thinking
THELMA: Task Based Holistic Evaluation of Large Language Model Applications-RAG Question AnsweringRetrieval-augmented generation → RAG robustness and adaptive retrieval
Benchmarking Spatiotemporal Reasoning in LLMs and Reasoning Models: Capabilities and ChallengesLLM reasoning → Spatial and embodied reasoning
Steering Risk Preferences in Large Language Models by Aligning Behavioral and Neural RepresentationsMechanistic interpretability → Activation steering methods
Signal in the Noise: Polysemantic Interference Transfers and Predicts Cross-Model InfluenceMechanistic interpretability → SAE features and their applications
Regularity and Stability Properties of Selective SSMs with Discontinuous GatingState-space sequence models (S4, Mamba) → Core SSM architectures for language
Stepwise Guided Policy Optimization: Coloring your Incorrect Reasoning in GRPOLLM reasoning → Reinforcement learning for reasoning
LLM Agents Are Hypersensitive to NudgesLLM agents and tool use → Social simulation and multi-agent debate
Towards Adaptive Categories: Dimensional Governance for Agentic AILLM agents and tool use → Agent threats, guardrails and governance
Physics Consistency and Latent Dynamics in Spatiotemporal Physics Field GenerationState-space sequence models (S4, Mamba) → Time series forecasting with SSMs
Quantization Meets Reasoning: Exploring and Mitigating Degradation of Low-Bit LLMs in Mathematical ReasoningLLM reasoning → Test-time compute and efficient thinking
HessFormer: Hessians at Foundation ScaleMixture-of-experts → Efficient MoE training and inference systems
Policy Gradient with Second Order MomentumDeep reinforcement learning methods → RL theory and sample complexity
QVGen: Pushing the Limit of Quantized Video Generative ModelsDiffusion models → Diffusion foundations and sampling
SoftCoT++: Test-Time Scaling with Soft Chain-of-Thought ReasoningLLM reasoning → Test-time compute and efficient thinking
Unsupervised Detection of Distribution Shift in Inverse Problems using Diffusion ModelsDiffusion models → Inverse problems and image restoration
Automatic Reward Shaping from Confounded Offline DataDeep reinforcement learning methods → Deep RL algorithms and representations
HelpSteer3-Preference: Open Human-Annotated Preference Data across Diverse Tasks and LanguagesLLM alignment / RLHF / preference optimisation → Reward modeling and RLHF
REACT: Runtime-Enabled Active Collision-avoidance Technique for Autonomous DrivingEnd-to-end autonomous driving → Adversarial scenarios and safety testing
Disentangling Reasoning and Knowledge in Medical Large Language ModelsLLM reasoning → Clinical and biomedical applications
HumaniBench: A Human-Centric Framework for Large Multimodal Models EvaluationMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
A Generative Framework for Causal Estimation via Importance-Weighted Diffusion DistillationDiffusion models → Diffusion foundations and sampling
GODBench: A Benchmark for Multimodal Large Language Models in Video Comment ArtMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
MegaScale-MoE: Large-Scale Communication-Efficient Training of Mixture-of-Experts Models in ProductionMixture-of-experts → Efficient MoE training and inference systems
When Thinking Fails: The Pitfalls of Reasoning for Instruction-Following in LLMsLLM reasoning → Prompting and chain-of-thought reasoning
MoE-CAP: Benchmarking Cost, Accuracy and Performance of Sparse Mixture-of-Experts SystemsMixture-of-experts → Efficient MoE training and inference systems
CARES: Comprehensive Evaluation of Safety and Adversarial Robustness in Medical LLMsLLM safety and security → Safety alignment, refusal and guardrails
Visual Planning: Let's Think Only with ImagesLLM reasoning → Spatial and embodied reasoning
EmotionHallucer: Evaluating Emotion Hallucinations in Multimodal Large Language ModelsMultimodal LLMs / vision-language models → Hallucination mitigation in LVLMs
Patho-R1: A Multimodal Reinforcement Learning-Based Pathology Expert ReasonerLLM reasoning → Clinical and biomedical applications +2
Can AI automatically analyze public opinion? A LLM agents-based agentic pipeline for timely public opinion analysisLLM agents and tool use → Social simulation and multi-agent debate
Diff-Unfolding: A Model-Based Score Learning Framework for Inverse ProblemsDiffusion models → Inverse problems and image restoration
LipDiffuser: Lip-to-Speech Generation with Conditional Diffusion ModelsDiffusion models → Audio, speech and music
GuideBench: Benchmarking Domain-Oriented Guideline Following for LLM AgentsLLM agents and tool use → General LLM agent frameworks and benchmarks
Learning Multimodal AI Algorithms for Amplifying Limited User Input into High-dimensional Control SpaceDeep reinforcement learning methods → RL for robot locomotion and manipulation
Benchmarking Critical Questions Generation: A Challenging Reasoning Task for Large Language ModelsLLM reasoning → Reasoning benchmarks and evaluation
Temporally-Grounded Language Generation: Towards Real-Time Vision-Language ModelsMultimodal LLMs / vision-language models → Video understanding with MLLMs
Decomposing stimulus-specific sensory neural information via diffusion modelsDiffusion models → Diffusion foundations and sampling
Reinforcement Learning Closures for Underresolved Partial Differential Equations using Synthetic DataDeep reinforcement learning methods → RL theory and sample complexity
Effective Probabilistic Time Series Forecasting with Fourier Adaptive Noise-Separated DiffusionDiffusion models → Flow matching, time series and science
Meta-World+: An Improved, Standardized, RL BenchmarkDeep reinforcement learning methods → Deep RL algorithms and representations
A Fourier Space Perspective on Diffusion ModelsDiffusion models → Diffusion foundations and sampling
Search and Refine During Think: Facilitating Knowledge Refinement for Improved Retrieval-Augmented ReasoningLLM reasoning → Knowledge- and tool-augmented reasoning +1
TCC-Bench: Benchmarking the Traditional Chinese Culture Understanding Capabilities of MLLMsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
SelfBudgeter: Adaptive Token Allocation for Efficient LLM ReasoningLLM reasoning → Test-time compute and efficient thinking
Semantic Caching of Contextual Summaries for Efficient Question-Answering with Language ModelsRetrieval-augmented generation → RAG robustness and adaptive retrieval
DRAGON: A Large-Scale Dataset of Realistic Images Generated by Diffusion ModelsDiffusion models → Image generation, text-to-image and editing
LD-Scene: LLM-Guided Diffusion for Controllable Generation of Adversarial Safety-Critical Driving ScenariosDiffusion models → Diffusion and flow policies for robotics +1
Diffusion-NPO: Negative Preference Optimization for Better Preference Aligned Generation of Diffusion ModelsDiffusion models → Diffusion and flow policies for robotics +1
Is PRM Necessary? Problem-Solving RL Implicitly Induces PRM Capability in LLMsLLM reasoning → Reinforcement learning for reasoning +1
HAPO: Training Language Models to Reason Concisely via History-Aware Policy OptimizationLLM reasoning → Reinforcement learning for reasoning
Sample Efficient Reinforcement Learning via Large Vision Language Model DistillationMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Unveiling the Potential of Vision-Language-Action Models with Open-Ended Multimodal InstructionsVision-language-action models → Generalist VLA models and benchmarks
GLOVA: Global and Local Variation-Aware Analog Circuit Design with Risk-Sensitive Reinforcement LearningDeep reinforcement learning methods → RL for energy, networks and operations
DiCo: Revitalizing ConvNets for Scalable and Efficient Diffusion ModelingDiffusion models → Diffusion foundations and sampling
FALCON: False-Negative Aware Learning of Contrastive Negatives in Vision-Language AlignmentMultimodal LLMs / vision-language models → Prompt learning and adaptation of CLIP
mmRAG: A Modular Benchmark for Retrieval-Augmented Generation over Text, Tables, and Knowledge GraphsRetrieval-augmented generation → RAG pipeline design and evaluation
CompAlign: Improving Compositional Text-to-Image Generation with a Complex Benchmark and Fine-Grained FeedbackDiffusion models → Image generation, text-to-image and editing
Real-Time Verification of Embodied Reasoning for Generative Skill AcquisitionLLM reasoning → Spatial and embodied reasoning
SoLoPO: Unlocking Long-Context Capabilities in LLMs via Short-to-Long Preference OptimizationLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Diffusion Models for Hyperspectral Image Analysis: A Comprehensive ReviewDiffusion models → Inverse problems and image restoration
MPMA: Preference Manipulation Attack Against Model Context ProtocolLLM agents and tool use → Agent threats, guardrails and governance
Bi-directional Recurrence Improves Transformer in Partially Observable Markov Decision ProcessesDeep reinforcement learning methods → Deep RL algorithms and representations
Human-Aligned Bench: Fine-Grained Assessment of Reasoning Ability in MLLMs vs. HumansLLM reasoning → Multimodal reasoning
Follow the Path: Reasoning over Knowledge Graph Paths to Improve Large Language Model FactualityLLM reasoning → Knowledge- and tool-augmented reasoning
Reinforcement Learning for AMR Charging Decisions: The Impact of Reward and Action Space DesignDeep reinforcement learning methods → RL for energy, networks and operations
Scalability of Reinforcement Learning Methods for Dispatching in Semiconductor Frontend Fabs: A Comparison of Open-Source Models with Real Industry DatasetsDeep reinforcement learning methods → RL for energy, networks and operations
One Image is Worth a Thousand Words: A Usability Preservable Text-Image Collaborative Erasing FrameworkDiffusion models → Robustness, privacy and adversarial aspects +1

All 505