Papers

New on arXiv

Mon 7 Sep 2026

410 papers

LatentPort: Beyond KV Cache - Cross-Model Transfer of Recurrent Memory in Hybrid Language Models: A 4B-to-9B Hybrid-State Handoff Without Target Prefix ReplayEfficient LLM inference and serving → KV cache optimization
CANAL: Channel-Aware Noise Allocation for Differentially Private Feature Distillation in Medical Image SegmentationKnowledge distillation → Distillation for medical imaging
Structure-Token Evidence-Anchored Reasoning for Scientific Chart UnderstandingLLM reasoning → Multimodal reasoning +1
An Empirical Measurement of Jailbreaking EvaluatorsLLM safety and security → Jailbreaking
Learning to Fly: Stable Vision-Guided UAV Servoing with Compact Target-Centric Cues and Reinforcement LearningDeep reinforcement learning methods → RL for robot navigation and UAV control
Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic TasksLLM agents and tool use → Agent skill learning and evolution
MI-PEFT: Mixture-of-Experts Integrated Parameter-Efficient Fine-Tuning Protein Language Models Improves Acidophilic Proteins ClassificationMixture-of-experts → Medical and biomedical MoE
VEX-Bench: Benchmarking LLM Agents for Assessing Exploitability of Software Supply Chain VulnerabilitiesLLM agents and tool use → Agent threats, guardrails and governance
Scaling Multi-Agent Systems with Prospect-State PropagationLLM agents and tool use → Social simulation and multi-agent debate
BanglaMemeX: Advancing Cultural Metaphoric Image Interpretation in Bangla with a Multimodal Explainable DatasetMultimodal LLMs / vision-language models → Social bias and fairness in VLMs
Eliciting Self-Verification in Multimodal Reasoning Agents with Reinforcement LearningLLM reasoning → Multimodal reasoning +1
What Does Multi-Agent LLM Debate Actually Change? A Layered Analysis of Disagreement and Answer QualityLLM agents and tool use → Social simulation and multi-agent debate
Mini-Batch Risk-Averse Deep Q-Learning: A Robot Navigation Case StudyDeep reinforcement learning methods → RL theory and sample complexity
HyCO: A Hybrid Neural Solver for Combinatorial OptimizationDiffusion models → Diffusion and flow policies for robotics +1
From Event Logs to Governed Action: A BlueSky Agenda for Agentic Process MiningLLM agents and tool use → Software engineering coding agents
Reasoning Beyond Transcription: Audio Language Models on Child Stuttering SpeechLLM reasoning → Audio-language model reasoning
MetaKV: Adaptive KV Cache Compression for Constrained LLM InferenceEfficient LLM inference and serving → KV cache optimization
TDDN: Text-aligned Diffused DINO Network for Puzzle UnderstandingLLM reasoning → Multimodal reasoning +1
FrogNano: Training a 4B Coding Agent via Online Task SynthesisLLM agents and tool use → Software engineering coding agents
JEDI: JEPA-to-Edge Distillation for Efficient Cropland Segmentation from Satellite ImageryKnowledge distillation → Detection and dense prediction distillation
PRIMUS: Identity, Governance, and Verification for Multi-Agent FederationsLLM agents and tool use → Agent threats, guardrails and governance
Deadline-Aware Adaptive Prefill Chunking for Efficient Large Language Model ServingEfficient LLM inference and serving → Serving systems and request scheduling
LLM Layers Immediately Correct Each OtherMechanistic interpretability → Knowledge and representations inside LLMs
InfluenceField: A Differentiable Field with Interventionally Identifiable Causal Structure for Multimodal World ModelingWorld models → World models for language agents +1
ComVLA: Communication-Aware Split Inference for VLA Models in 6G-Connected RoboticsVision-language-action models → Efficient VLA inference and adaptation
SAFIRE: Safety-Critical Benchmark for Fine-grained Fire and Smoke Understanding in Multimodal LLMsMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness
A*-Thought-V2: Efficient Latent Reasoning via Geometric Dynamics of LLMLLM reasoning → Test-time compute and efficient thinking
VoT: Vision-of-Thought for Unified Multimodal Representation AlignmentLLM reasoning → Multimodal reasoning
Latent-MoE: Domain-Aware Mixture-of-Experts for PDEs with Multi-Regime PhysicsMixture-of-experts → MoE theory and gating functions
You Can't Prefer Emotions You Don't Sample: Intensity Undershoot in DPO-Tuned LLMsLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
LLM Agents as Computational TypologistsLLM agents and tool use → General LLM agent frameworks and benchmarks
Signed Rescue Routing: Harm-Aware Cascades for Efficient LLM InferenceEfficient LLM inference and serving → Serving systems and request scheduling
What Does an LLM-Agent Leaderboard Rank Actually Compare?LLM agents and tool use → General LLM agent frameworks and benchmarks
Decomposition-Guided Diffusion Language Models for Inertial Confinement Fusion PredictionDiffusion models → Flow matching, time series and science
LLM Forensics: Where Do Backdoors Hide? Localizing and Controlling Trigger Mechanisms with Sparse AutoencodersLLM safety and security → Backdoor and data poisoning attacks +1
The Emerging AI Paper-Review Arms Race: Adversarial Co-Evolution in Scholarly PublishingLLM agents and tool use → Web, search, and research agents
APPSim-Bench: Bridging Real-world Apps and Reproducible Evaluation for Mobile GUI AgentsLLM agents and tool use → GUI and mobile agents
ParetoTransport: Generative Optimization by Mass Transport Toward The Pareto FrontDiffusion models → Diffusion foundations and sampling
On the Recall Scaling Laws in Mamba: A Theoretical and Mechanistic Study via HashingState-space sequence models (S4, Mamba) → Core SSM architectures for language
Audit Without Verification: When LLM Accountability Layers Relay Rather Than CheckLLM agents and tool use → Agent threats, guardrails and governance
AgentIdeaBench: Benchmarking Scientific Ideation in the Agent EraLLM agents and tool use → Web, search, and research agents
Search-to-World: Evaluation of 3D World Delivery from User Request through Web SearchLLM agents and tool use → Web, search, and research agents
I Don't Miss You, but I Do: Self-Explanation Faithfulness of Modality Missingness in Vision-Language ModelsMultimodal LLMs / vision-language models → VLM benchmarks and visual reasoning
ICI-VLA: In-Context Imitation with Spatiotemporally Aligned Demonstrations for Vision-Language-Action ModelsVision-language-action models → Generalist VLA models and benchmarks
From Simulated Citizens to Simulated Deliberation: Challenges in Representation and InteractionLLM agents and tool use → Social simulation and multi-agent debate
Qwen-Audio-3.0-ASR Technical ReportMixture-of-experts → Multimodal MoE models
Analytical Resource Management for Fine-grained MoE Computation-Communication OverlapMixture-of-experts → Efficient MoE training and inference systems
PhysReal: Learning Real-World Deformable Object Physics via Hybrid Constitutive ModelingWorld models → Latent predictive world models
CoER: Defending against Adaptive Indirect Prompt Injection via Adversarial Co-Evolution and RefinementLLM agents and tool use → Agent threats, guardrails and governance +1
Topologically Consistent Agricultural Parcel Vectorization with Semantic-Guided Diffusion and Topology-Aware PolygonizationDiffusion models → Inverse problems and image restoration
MEMO: Multimodal Evidence Memory Organization for Long-Horizon LLM AgentsLLM agents and tool use → Agent memory systems
Measuring Language Transfer in Robot Policies: Adding Greek to a Cosmos3 Vision-Language-Action PolicyVision-language-action models → Generalist VLA models and benchmarks
Temporal-Causal Inference for Reinforcement Learning via Automata LearningDeep reinforcement learning methods → Deep RL algorithms and representations
Parser-Free VLM Verification for Federated Weakly Supervised Video Anomaly DetectionMultimodal LLMs / vision-language models → Chart understanding and anomaly detection
Unified Vision-Centric Pedestrian Crossing Action Prediction via Adaptive Patch Projection and Proactive Spatial RectificationMultimodal LLMs / vision-language models → Driving VLMs and open-vocabulary detection
Think Wider: Mitigating Latent Rank Collapse in Implicit Chain-of-Thought ReasoningLLM reasoning → Prompting and chain-of-thought reasoning
OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model PretrainingWorld models → Embodied and robot world models
Uncertainty Quantification for LLM Agents: A Taxonomy, an Evaluation Protocol, and an Empirical StudyLLM agents and tool use → General LLM agent frameworks and benchmarks
Social Intuition vs. Machine Reasoning: Anticipating Human-Robot Interaction from multiple modalitiesLLM reasoning → Spatial and embodied reasoning +1
Beyond Single-Negative Preference: Multi-Negative DPO for LLM-Centric Historical Entity LinkingLLM alignment / RLHF / preference optimisation → LLM preference optimization methods
Beyond Fluent Generation: A CPU Reliability Benchmark for MCP-Style Tool Calling in Sub-2B Small Language Models for Edge DeploymentLLM agents and tool use → Tool learning and function calling
Scanning the Harness: A Repository Study of Configuration Exposures in AI Coding AgentsLLM agents and tool use → Software engineering coding agents
AAS-RAIL: Improving Information Extraction for Asset Administration Shells through Retrieval-Augmented In-Context LearningRetrieval-augmented generation → RAG pipeline design and evaluation
PV-WM: A Heterogeneous Micro-Macro World Model for Articulated Pedestrian-Vehicle Co-RolloutWorld models → World models for autonomous driving
PCFlow: Physics-Conditioned Flow Matching for GPR B-Scan Image SynthesisDiffusion models → Flow matching, time series and science
World Models Under Asynchronous Sensor ObservationsWorld models → Latent predictive world models
Merging Cyber Threat Intelligence Through Retrieval-Augmented Generation and Small Language Models for Rich Threat RepresentationRetrieval-augmented generation → Privacy and security of RAG
Matryoshka Hash Representations for Model-Aware Compact Semantic RetrievalRetrieval-augmented generation → RAG pipeline design and evaluation
MV-STRIDE: Enabling MLLMs to Master Multi-View Spatial Reasoning via Hierarchical Capability ModelingLLM reasoning → Spatial and embodied reasoning +1
SkillAlign: Aligning Skill Interfaces for LLM-based AgentsLLM agents and tool use → Agent skill learning and evolution
Elastic Horizon: Discovering the Effective Interaction Frontier in Agentic Reinforcement LearningLLM agents and tool use → Reinforcement learning for LLM agents
Proximity-CLIP: Text-Guided Semantic Proximity Learning for Zero-Shot Anomaly DetectionMultimodal LLMs / vision-language models → Chart understanding and anomaly detection
Distance-Aware Attention and Wall-Distance Expert Routing for Transformer-Based 3D Flow PredictionMixture-of-experts → Vision mixture-of-experts
The Price of Consistency: Exploiting Visual Anchors for Multimodal Jailbreaking in Video GenerationLLM safety and security → Vision-language and text-to-image safety
Agentic Algorithm Engineering: Improving Shared-Memory Exact Minimum CutsLLM agents and tool use → General LLM agent frameworks and benchmarks
Enabling High-Bandwidth Flash for Generative Recommendation Serving with Write-Aware KV Cache PolicyEfficient LLM inference and serving → KV cache optimization
PhysMAS: Physics-Grounded Multi-Agent Synthesis of Compositional 4D GaussiansLLM agents and tool use → Multimodal and embodied agents
In-Place Instruction Following in Diffusion Language ModelsDiffusion models → Discrete and text diffusion
Stable-MM-R1: Anchoring Multimodal Reasoning Dynamics via Entropy-Guided StratificationLLM reasoning → Reinforcement learning for reasoning
Flow3D-OPD: Multi-Teacher On-Policy Distillation for 3D Geometry Generation with Flow-Matching Diffusion TransformerDiffusion models → 3D, point cloud and scene generation +1
Retrieval-Augmented Multi-Prompt Ensemble for Minor-Grain Breeding Information ExtractionRetrieval-augmented generation → Time series and agricultural RAG
AgentLeak: Cloning Stronger LLM Agent Capabilities onto Weaker Agents Beyond Skill StealingLLM agents and tool use → Agent threats, guardrails and governance
Beyond Task Success: Stage-Wise Reliability of World Model Planning under Sensing DegradationWorld models → Latent predictive world models
Online Draft Co-Training for Speculative Decoding in Large-Scale, Long-Context RL Post-TrainingEfficient LLM inference and serving → Speculative decoding +1
Revisiting Complete Reasoning Traces for Post-TrainingLLM reasoning → Test-time compute and efficient thinking
RoboDreamer: Anticipatory Humanoid Locomotion with Predictive State-Space ModelsState-space sequence models (S4, Mamba) → Video, motion and generative Mamba models
CASCADE: A Spatio-Temporal-Causal Reasoning Representation and Dataset for DrivingLLM reasoning → Chain-of-thought reasoning for driving
Where to Look and What to Use: Retrieve-Localize-Generate for Long-Term Conversational Memory Question AnsweringRetrieval-augmented generation → Conversational and speech RAG
Single Image to Textured 3D Object Generation in Frequency Domain: From Theory to PipelineDiffusion models → 3D, point cloud and scene generation
SupGRPO: Enhancing GRPO with Matching-based Online SFT for Text SpottingLLM reasoning → Reinforcement learning for reasoning
A Hierarchical Consistency Framework for Auditing Retrieval-Augmented Generation SystemsRetrieval-augmented generation → RAG robustness and adaptive retrieval
VST: Verifiable Structured Transport for Auditable Agent-to-Agent Alpha DiscoveryLLM agents and tool use → Agent memory systems
SpatialBlock: Enhancing Spatial Intelligence in LVLMs via Synthetic Block-Stacking ProblemMultimodal LLMs / vision-language models → Embodied and 3D vision-language models
Trust-But-Verify: Poisoning-Resilient Locally Private Graph Learning ProtocolsLLM safety and security → Membership inference attacks
PhysSAE: Mechanistic Interpretability of PINNs with Sparse AutoencodersMechanistic interpretability → Sparse autoencoder training and evaluation
TrojanWorld: Backdooring World-Model Agents via Imagination SteeringWorld models → World model security and GUI agents +2
Beyond One-Shot Expansion: Contrastive Evidence Exploration for Multi-Hop RetrievalRetrieval-augmented generation → Knowledge-graph and structured-knowledge RAG
FreqDoor: A Hidden Trojan in the Frequency Domain for Backdoor Attacks on Vision-Language ModelsMultimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness +1
MEMOBench: A Process Level Memory Benchmark for Robotic ManipulationVision-language-action models → Reasoning and chain-of-thought in VLA
Disentangling Steering VectorsMechanistic interpretability → Activation steering methods

All 410