Papers

New on arXiv

Mon 24 Aug 2026

491 papers

Spectral-Guided Diffusion: Accelerating Inference via Static Spectral Layer SchedulingDiffusion models → Diffusion foundations and sampling
Task-Aware Spectral Pruning: A Mixture-of-Masks Framework for Efficient LLM InferenceEfficient LLM inference and serving → Early exit and edge inference
Read-Best Is Not Steer-Best: A Probing--Steering Layer Dissociation in Omni-Modal Large Language ModelsMechanistic interpretability → Activation steering methods
Robust Federated Q-Learning with Almost No CommunicationDeep reinforcement learning methods → RL for energy, networks and operations
The Illusion of Independent Quorums: Epistemic Fault Domains and Correlated Cognitive Failures in Agentic QuorumsLLM agents and tool use → Agent memory systems
Memory-Efficient Training-Free Acceleration of Diffusion Transformers with BaryCacheDiffusion models → Diffusion foundations and sampling
A Reranker for Orchestrating Heterogeneous Speech and Text RetrieversRetrieval-augmented generation → Generation and context integration
AffectOmni: RL-Verifiable People-Centric Grounded Affective Reasoning for Social and Art-Related ScenesLLM reasoning → Audio-language model reasoning +1
Comparing Chunking and Embedding Strategies for Turkish RAG SystemsRetrieval-augmented generation → Multimodal RAG
The Dialect Tax: Dialectal Biases Persist throughout the Language Modeling PipelineMechanistic interpretability → Knowledge and representations inside LLMs
Demystifying Reinforcement Learning Post-Training of Language ModelsLLM reasoning → Reinforcement learning for reasoning
MacroAgent: Regularity-Aware Macro Legalization with LLM-Agent-Designed Contour AlgorithmsLLM agents and tool use → Software engineering coding agents
ExFold: Unified Expert Folding for Training-Free MoE Prefill-Decode AccelerationMixture-of-experts → Efficient MoE training and inference systems
HAP: Head-Adaptive Visual Token Pruning via Cross-Modal AlignmentMultimodal LLMs / vision-language models → Visual token pruning and efficiency +1
The Loss Floor of Denoising Score Matching: Fisher Geometry from Schr"odinger BridgesDiffusion models → Diffusion foundations and sampling
PROOF-Gen: From Optimized Data to Better DistillationLLM agents and tool use → Tool learning and function calling
Retrieval-augmented generation vs. deterministic tax computation in multi-agent financial advisory: A 2x2 factorial experimentRetrieval-augmented generation → Legal and financial RAG
BenchBench-Protocol: Evaluating Real-World Wet-Lab Protocol Reasoning and ModificationLLM reasoning → Clinical and biomedical applications
LG-GER: Language-Guided Group Emotion Recognition via Multimodal Evidence DistillationMultimodal LLMs / vision-language models → Multimodal emotion understanding
Semantic Overlays: Mitigating Prompt Injection with Annotations Beyond Tokens and Steering VectorsLLM safety and security → Prompt injection and agent security +1
Markets, Not Planners: Decentralized Orchestration of LLM Agents with Private InformationLLM agents and tool use → General LLM agent frameworks and benchmarks
DreamLedger: Where to Refuse World-Model Imagination Using Execution-Settled CreditWorld models → Embodied and robot world models
Beyond the Mandate: A Systematic Security Analysis of the Agent Payments Protocol (AP2)LLM agents and tool use → Agent threats, guardrails and governance
Does Episodic Memory Help Close the Lexical Frequency Gap in Sensitivity to Syntactic Contrasts? A Test Using Retrieval-Augmented Language ModelsRetrieval-augmented generation → RAG robustness and adaptive retrieval
Exploit More, Explore Smarter for Budget-Constrained Agentic SearchLLM agents and tool use → General LLM agent frameworks and benchmarks
PuzzleKV: Page-Wise Low-Rank Decomposition for KV Cache CompressionEfficient LLM inference and serving → KV cache optimization
Pipeline-Native Transformers: Co-Designing Model Architecture and CPU Inference for Bandwidth-Efficient Autoregressive DecodeMixture-of-experts → Efficient MoE training and inference systems
Minima-KV: Retention-Preserving KV Cache Compression with Mixed-Format Paged AttentionEfficient LLM inference and serving → KV cache optimization
Learning to Act While Waiting: RL Finetuning of Generalist Robot Policies Under Inference LatencyVision-language-action models → Reinforcement learning and post-training +1
Generating Biomedical Fact-Checking Reports with RL-Enhanced Agentic SearchLLM agents and tool use → Medical and scientific domain agents
Discovering Cross-Language Reasoning Invariance in LLMs with Geometry-Invariant Sparse AutoencodersLLM reasoning → Reasoning benchmarks and evaluation +1
Serving Masked Diffusion LLMs: Characterization and Design Principles from Real HardwareDiffusion models → Discrete and text diffusion
Giga-Embeddings: Mixture-of-Experts Encoders for High-Throughput Text EmbeddingsMixture-of-experts → MoE large language models
A Theory of Speciation in Generative Diffusion Models on Compact Riemannian ManifoldsDiffusion models → Diffusion foundations and sampling
Mixture of Channel Experts: Static Sparse Supports with Input-Adaptive Mixing for Pointwise ProjectionsMixture-of-experts → Vision mixture-of-experts
Primate vision reveals a missing principle for robust dynamic AIWorld models → Embodied and robot world models
TrustShiftProbe: Characterizing, Benchmarking, and Defending Staged Trust Attacks on MCP ServersLLM agents and tool use → Agent threats, guardrails and governance
Too much of a good thing -- when knowledge distillation promotes overfitting, and how to avoid itKnowledge distillation → General knowledge distillation methods
AgentRoom: Concurrent Multi-Agent Coding in a CRDT-Backed Shared WorkspaceLLM agents and tool use → Software engineering coding agents
DriftAD: Visually-Guided Text Drift for Few-Shot Industrial Anomaly DetectionMultimodal LLMs / vision-language models → Chart understanding and anomaly detection
Platonic Representation Hypothesis on World ModelsWorld models → Latent predictive world models
ADE: Agentic Data Evolution Framework for Human-Centered ObjectivesLLM agents and tool use → General LLM agent frameworks and benchmarks +1
Do LLMs Understand Limit Order Book Dynamics?World models → World models for language agents
Renormalization Group Flow Matching for Scalable Local Generative ModelingDiffusion models → Flow matching, time series and science
Automata from Agent Traces: Failure and Next-Step PredictionLLM agents and tool use → Software engineering coding agents
Gated Activation Steering for Reducing Sycophancy & Hallucination in Medical Question AnsweringMechanistic interpretability → Hallucination detection from model internals
Scaling Reinforcement Learning for Diffusion Models via Velocity MatchingDiffusion models → Diffusion and flow policies for robotics +1
Confidently Wrong, Silently So: Auditing Undetectable Failures of a Deployed On-Device Language ModelLLM safety and security → Safety alignment, refusal and guardrails
Elastic KV Cache for LLM Serving:A Working Reclamation Mechanism, and Why Chunked Prefill Already Closes the GapEfficient LLM inference and serving → KV cache optimization
Beyond Executable Models: The Pufibara Agent Harness and the Modelica Agent Workflow Benchmark for Physical System ModelingLLM agents and tool use → Software engineering coding agents
Feedback That Backfires: Why Small Language Model Agents Repeat the Call They Just Watched FailLLM agents and tool use → Tool learning and function calling
MolEmb: Multimodal Large Language Models Can Be Strong Molecular Embedding ModelsMultimodal LLMs / vision-language models → Chart understanding and anomaly detection
Best Practice Critic OptimizationLLM reasoning → Reinforcement learning for reasoning
ReWorld: An Interactive World Model with Long-Horizon MemoryWorld models → Video-generation world models and simulators
SWE Refactor Bench: Can Coding Agents Complete a Long-Horizon, Whole-Repository Stack Migration?LLM agents and tool use → Software engineering coding agents
Provably adaptive sampling with uniform and remasking discrete diffusion modelsDiffusion models → Diffusion foundations and sampling
Prime Agent: A Self-Improving RLM HarnessLLM agents and tool use → Software engineering coding agents +1
ConvergeFlow: Language Flow with Provable Convergence to Token EmbeddingsDiffusion models → Discrete and text diffusion
The Interaction Tax: When Communication Erases Diversity in Multi-Agent TeamsLLM agents and tool use → Social simulation and multi-agent debate
Interpretable AI with Local DistillationKnowledge distillation → General knowledge distillation methods
Correcting a learned physical invariant improves world-model rolloutsWorld models → Latent predictive world models
Investigating Relational Reasoning in VLMsLLM reasoning → Multimodal reasoning +1
An Interactive Agent for Requirement-Driven Candidate SourcingLLM agents and tool use → Software engineering coding agents
SVD-Based Typicality Maps for Out-of-Distribution Detection in Vision TransformersMechanistic interpretability → Interpreting vision and multimodal models
Mitigating Reasoning-Induced Misalignment via Safety-Direction PenaltyLLM reasoning → Reasoning for agents, planning and code +1
SRPO: Self-Reflective Policy Optimization for Long-Horizon ReasoningLLM reasoning → Test-time compute and efficient thinking
GeoWAM: Visual Geometry World Action Models for Autonomous DrivingWorld models → World models for autonomous driving +1
Act with Intent: Distilling Behavior Intent for Vision-Language-Action ModelsVision-language-action models → VLA architectures and generalization
What's the Catch? Evaluating Temporal Consistency in Vision-Language ModelsMultimodal LLMs / vision-language models → Video understanding with MLLMs
MetaCaster: Meta-Harness-Optimized Agent for End-to-End Few-Shot Learning of Lightweight Time Series ForecastersLLM agents and tool use → Financial trading and forecasting agents
InjecMEM: Memory Injection Attack on LLM Agent Memory SystemsLLM agents and tool use → Agent threats, guardrails and governance
Reward-Free Continual Adaptation for Resilient Space RobotsWorld models → Latent dynamics models and model-based RL
Temporal Property-driven Design Space Exploration with Reinforcement Learning for Cyber-Physical SystemsDeep reinforcement learning methods → Safe and constrained RL
Image-Conditioned Diffusion Models for Quality Assurance of Organ-at-Risk Segmentations in RadiotherapyDiffusion models → Medical imaging and segmentation
ChebBooster: A Training-Free Approach for Efficient Diffusion Transformer Inference via Chebyshev-Inspired ExtrapolationDiffusion models → Diffusion foundations and sampling
MomADv2: Reliable Temporal Memory for End-to-End Autonomous DrivingEnd-to-end autonomous driving → End-to-end system architecture
Adaptive Item-based Collaborative Structures via Noise Rescheduling in Diffusion for Generative RecommendationDiffusion models → Diffusion recommenders
Right-Sizing LLM-Agent Decomposition in VAT Determination: A Pilot Controlled SweepLLM agents and tool use → Financial trading and forecasting agents
Long-Horizon Audio-Visual Generation for Persistent Stories and Interactive WorldsWorld models → Video-generation world models and simulators
DF-MoE: Generalizable Deepfake Detection via Multimodal Sparse Mixture-of-ExpertsMixture-of-experts → Multimodal MoE models
OptiSight: Bridging Semantic Reasoning and Geometric Control for Embodied NavigationLLM reasoning → Spatial and embodied reasoning
Controllable blind deblurring with diffusion modelsDiffusion models → Inverse problems and image restoration
DPIAgent: Divide, Protocol, Isolate for Agentic Reproduction Test GenerationLLM agents and tool use → Software engineering coding agents
Can Coding Agents Build Robust Baselines? A Skill-Based Approach for Automating the Medical Imaging Model-Development PipelineLLM agents and tool use → Medical and scientific domain agents
IntentQA: Intent Question Answering in Videos by Cognitive Context ReasoningLLM reasoning → Multimodal reasoning
ROS2SmolVLA: Enabling Small Vision-Language-Action Models for Integration into Industrial-Grade Lightweight RobotsVision-language-action models → Generalist VLA models and benchmarks
Agent-G2: Gaussian Guidance for Agentic Reinforcement LearningLLM agents and tool use → Reinforcement learning for LLM agents
EviSafe: Evidence-Grounded Safety Evaluation for Vision-Language ModelsMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness
Sigmoid Attention as a Better Substrate for Learned KV Cache EvictionEfficient LLM inference and serving → KV cache optimization
Apodex 1.1: Scaling Agentic Intelligence for Complex WorkLLM agents and tool use → General LLM agent frameworks and benchmarks
From Natural Language Policies to Executable Obligations: A Verification Harness for Dependable In-Car LLM AgentsLLM agents and tool use → Agent threats, guardrails and governance
Spatiotemporally Decoupled Autoregressive Diffusion Model for Human Motion GenerationDiffusion models → Video generation
Dual-Grained Agent Memory and Shapley Context Attribution for Multimodal Agentic LearnerLLM reasoning → Reasoning for agents, planning and code +2
Is Next-Chunk Reasoning RL Really Better than SFT? Revisiting Training Strategies under no-CoT DataLLM reasoning → Prompting and chain-of-thought reasoning
E2S-Pruner: Progressive Two-Stage Evidence Fusion for Visual Token Pruning in Vision-Language ModelsMultimodal LLMs / vision-language models → Visual token pruning and efficiency +1
The Laws of Context Allocation: Causal Measurement and Closed-Loop Orchestration in Generative SearchRetrieval-augmented generation → RAG robustness and adaptive retrieval
Mind the Couch! Eliciting MLLM Reasoning in Interior Design via Weak-to-Strong Task Vector InjectionMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
Retrieval-Augmented Classification of Environmental Mitigations in Hydropower Licensing DocumentsRetrieval-augmented generation → Agentic RAG, agents and memory
MLLM-Assisted Audio VOS: A 3rd Place Report for the MeViS-Audio Track, 8th LSVOS ChallengeMultimodal LLMs / vision-language models → Video understanding with MLLMs
Think Only When Needed: Prompt-Authority Control for Selective Slow-Path Intervention in Vision-Language-Action ManipulationVision-language-action models → Reasoning and chain-of-thought in VLA

All 491