AI 论文 · 2026-09
浏览 2026-09 发布的AI 论文内容，第 7 页，共 741 条。
- DianShi-RxnDB: A Large-Scale, Fine-Grained Organic Reaction Data Platform Built via a Fully Automated Pipeline for Researchers and AI Agents
- PARSER: Read in Parallel, Reason in Depth for Long-Context LLM Agents
- TransNormal-2: Geometry-Grounded Rectified Flow with Edge-Aware Decoding for Precise Normal Estimation
- VidaForge: Open Research Infrastructure for Video Pretraining Data Recipes
- OracleZoom: On-Policy Self-Distillation Inspired Reference-Constrained Recursive Image Super Resolution
- Multi-Grid Post-Training for Long-Form Multi-Shot Video Generation
- Adaptive Bridge: A Proxy-Based Decoupling Layer for Mitigating DDS Backpressure in ROS 2
- MLLMs Hallucinate when Information Distribution Drifts in Synergy Heads
- Steering Geometry: Validating Human Value Geometry in LLM Steering Space
- MobileVLA-R1 2.0: RL-Enhanced Reasoning for Mobile Robot Control
- VDiff-Bench: A Challenging Benchmark for Fine-Grained Image Difference Identification
- Counter-Swarm Doctrine: Containing Coordinated Agent Intrusions
- DataFlex-RL: An Evaluation Platform for RLVR Data Policies
- DriveZero: End-to-End Driving Beyond Human Demonstrations
- SkillSpec: Intent-Masked Specification Reasoning for Agent Skill Correctness
- Tri-PvP: Exposing Modality Bias in Omni-Modal Large Language Models through Perceptual-Propositional Evidence Conflicts
- Cadence: Error-Bounded Lossy Compression of Demand Time Series with a Time-Series Foundation Model
- EvoSafeHarness: Evolving Model- and Domain-Specific Harnesses for Securing Agents
- Beyond Top-k Skill Retrieval: Diversity-Aware Skill Routing for LLM Agents
- Online Learning with LLM Experts from Limited Feedback
- Diffs vs. Whole Files: An Empirical Comparison of Iterative Edit-Based and Direct Generation for Flutter/Dart Code Models
- Occamy-1.0: Open Pareto-frontier 35B Intelligence for Co-work
- RenderFormer-V2: Neural Rendering with Heterogeneous Scene Primitives
- What LLM Trading Agents Actually Do in Production: A Six-Month, Population-Scale Record from Two Fleets
- Srijika: OpenType-Layout-Reusing Font Restyling for Nine Indic Scripts
- SceneMosaic: Efficient and Diverse Simulation-Ready Scene Generation via Hybrid Agentic Layout Evolution
- What Did I Just Say? Self-Listening for Full-Duplex Speech Models
- GE-Act 2.0: Pretraining and Scaling a World-Action Model for Robotic Manipulation
- Grounded Skill Synthesis from Code at Scale for Agentic Intelligence
- WorldSculpt: Generating Compositional Worlds from Grounded Videos
- UniMate: One Unified Model to Animate Diverse Skeletons
- WearableQA: A Benchmark for Health Reasoning over Real-World Wearable Data
- RoboSPA: Can VLA Models Go Beyond Simple Scenes and Short-Horizon Tasks?
- RISE: Recursive Improvement via Self-Extrapolating Policy Distillation
- Don't Drop Dropout: Optimizing Layer Sparsity for Efficient LLM Training and Inference
- Ask Before You Optimize: Dynamic Pre-Formulation Clarification for Interactive Optimization
- BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference
- Learning 3D Editing without Paired Supervision via Generative Prior Distillation
- Beneath the Surface of Chains-of-Thought: A Mechanistic Interpretation of Reasoning Operations in LLMs
- Knowing What Not to Answer: Selective Non-Compliance in Vision-Language Models
- Refuse without Refusal: A Structural Analysis of Safety-Tuning Responses for Reducing False Refusals in Language Models
- τ^τ-Bench: An Environment for End-To-End, Realistic Agent Construction
- Not All Ranks Are Equal: Budget-Aware LoRA Merging Across Tasks
- MaxKernel: Agentic Kernel Generation for TPUs
- When Quantization Breaks Memory: Recurrent-State Write-Back in Low-Precision Temporal Inference
- Safety for Whom? Boundary-Aware Self-Distillation for Controlled LLM Safety Refusal
- HarvestBench: Measuring Whether LLM Agents Will Pay to Avoid Killing Animals
- Privacy Failure in Split-LLM Training, The Returned Gradient Nullifies the Decoys
- AdaptVPR: Route-Aware Hard Positive Generation for Robust Visual Place Recognition
- Iris: Climbing to the Search Frontier
- EVOHARNESSBENCH: Can Your Agents Keep Pace with an Evolving Harness?
- Scal3R: Learning Efficient Multi-Relative Pose Query for Scalable Online 3D Reconstruction
- Principia: Relational Physics Tests for Video Models
- Compile by Training: Turning Natural-Language Specifications into Local Neural Functions
- Puffin-World: Scaling a Unified Multimodal Model with Native 3D World States
- One Editor, Many Edits: A Unified Training-Free Framework for Diverse Video Editing
- Last Translation Benchmark
- Rethinking On-Policy Distillation of Large Language Models II: One Training Example
- Terminal-Universe: Turning Agent Trajectories into Scalable Terminal Environments
- Beyond Retrieval: Progressive Latent Memory Evolution for Streaming Video Understanding
- Environment Evolution for Terminal Agents
- Why Gated DeltaNet Survives 4-Bit Quantization: NVFP4 W4A4 for the Recurrent Half of a Hybrid 27B LLM
- DRACO: Fine-Grained Credit Assignment with Dynamic Rubrics for Long-Horizon Agent Training
- CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation
- When Models Edit Too Much: On the Fidelity of Minimal Code Edits
- Editable Visual Design
- Unlocking Lossless Speedups in LLMs via Discrete Diffusion
- WorldReward: Reward Modeling for Camera-Conditioned World Models
- Select, Compress, Reinvest: A Controlled Study of Visual-Token Allocation in Long-Video MLLMs
- LLaDA-Image: Building Strong Image Generators with Fully Open Training Recipes
- ENEAS: Embedding-guided Neural Ensemble for Adaptive Segmentation
- Unfold The World: Factorize 4D Properties in Reinforcing Spatial Reasoning
- How Far Can Synthetic Data Take Thai OCR?
- The Attention Triangle in Audio-Video Models
- FlashRender: Few-Step Generative Rendering via Camera-Controlled Video MeanFlow
- Building and Evaluating Fixed-Voice Thai TTS from Synthetic Speech
- Random Attention: Rethinking KV Cache Eviction for Efficient Reasoning
- PACE: Towards Surfacing Hidden Conflicts in User Requests
- What Else Needs Fixing? Exploring Cost-Effective Test-Time Compute for Revision Propagation in Artifacts Generated Through Conversation
- FlowBalance: Verifier-Grounded Self-Improvement from On-Policy Reasoning Experience
- The 2026 PNPL Competition: Word Classification and Efficient Cross-Subject Generalisation in LibriBrain100
- Measuring the Checker: Mutation Analysis for GPU-Kernel Benchmark Oracles
- ShieldVLA: Feasibility-Aware Safety Alignment for Vision-Language-Action Models
- SimpleMemVLA: A Simple but Effective Native-Video Memory for Vision-Language-Action Models
- MasterControl Seventeen Every Time
- RoboTok: An Internet-Scale Data Engine for Human Demonstration Retrieval and Dexterous Manipulation Learning
- VeriPhy: Agentic Physical Reasoning for World Model Evaluation and Refinement
- Unifying Conformal Language Tasks with In-Context Ensembles
- Causal Foundation Models
- Verify Before You Distill: Prompt-Level Teacher Gating for On-Policy Distillation
- A Common Measure of Communication for Speech Brain-Computer Interfaces
- SolarWM: Open Data and Scalable Training for Long-Horizon Video World Models
- Graph Machine: Towards Better Pretraining via Edges
- Post-Training Language Models for Gold-Medal Performance in Coding Competitions
- Cliff: Learning Process Rewards from the First Mistake
- VibeVoice-ASR-Streaming Technical Report
- EarlyEval: Cheaper Agent Evaluation via Early Outcome Prediction
- ShallowStream: Index Shallow then Answer Deep for Streaming Video Understanding
- From Reweighting to Rewriting: Unlocking the Intervention Effects of Influential Samples in Training Data Attribution
- Bilevel Coordinated Reflection: A Game-Theoretic Approach to Multi-Agent LLM Systems