arXiv cs.AI 周报
人工智能领域最新研究进展
📊 研究方向热度分析
🤖 LLM Agents
193篇
该研究方向持续发展，产出多篇高质量论文。
- SWE-QA-Pro: A Representative Benchmark and Scalable Training Recipe for Repository-Level Code Understanding
- Proactive Rejection and Grounded Execution: A Dual-Stage Intent Analysis Paradigm for Safe and Efficient AIoT Smart Homes
- Symphony: A Cognitively-Inspired Multi-Agent System for Long-Video Understanding
- Graph-Native Cognitive Memory for AI Agents: Formal Belief Revision Semantics for Versioned Memory Architectures
- AI Scientist via Synthetic Task Scaling
🧠 Reasoning
191篇
该研究方向持续发展，产出多篇高质量论文。
- Sample-Efficient Adaptation of Drug-Response Models to Patient Tumors under Strong Biological Domain Shift
- ASDA: Automated Skill Distillation and Adaptation for Financial Reasoning
- NeSy-Route: A Neuro-Symbolic Benchmark for Constrained Route Planning in Remote Sensing
- Can LLMs Model Incorrect Student Reasoning? A Case Study on Distractor Generation
- Empowering Chemical Structures with Biological Insights for Scalable Phenotypic Virtual Screening
🛡️ Safety & Alignment
147篇
该研究方向持续发展，产出多篇高质量论文。
- Empowering Chemical Structures with Biological Insights for Scalable Phenotypic Virtual Screening
- AnoleVLA: Lightweight Vision-Language-Action Model with Deep State Space Models for Mobile Manipulation
- Detecting the Machine: A Comprehensive Benchmark of AI-Generated Text Detectors Across Architectures, Domains, and Adversarial Conditions
- Security Assessment and Mitigation Strategies for Large Language Models: A Comprehensive Defensive Framework
- Do Understanding and Generation Fight? A Diagnostic Study of DPO for Unified Multimodal Models
⚡ Efficiency
146篇
该研究方向持续发展，产出多篇高质量论文。
- Sample-Efficient Adaptation of Drug-Response Models to Patient Tumors under Strong Biological Domain Shift
- ASDA: Automated Skill Distillation and Adaptation for Financial Reasoning
- Frequency Matters: Fast Model-Agnostic Data Curation for Pruning and Quantization
- Proactive Rejection and Grounded Execution: A Dual-Stage Intent Analysis Paradigm for Safe and Efficient AIoT Smart Homes
- Empowering Chemical Structures with Biological Insights for Scalable Phenotypic Virtual Screening
🎮 Reinforcement Learning
144篇
该研究方向持续发展，产出多篇高质量论文。
- HIPO: Instruction Hierarchy via Constrained Reinforcement Learning
- How Log-Barrier Helps Exploration in Policy Optimization
- From Digital Twins to World Models:Opportunities, Challenges, and Applications for Mobile Edge General Intelligence
- CLAG: Adaptive Memory Organization via Agent-Driven Clustering for Small Language Model Agents
- Resilience Meets Autonomy: Governing Embodied AI in Critical Infrastructure
👁️ Vision-Language
138篇
该研究方向持续发展，产出多篇高质量论文。
- NeSy-Route: A Neuro-Symbolic Benchmark for Constrained Route Planning in Remote Sensing
- AnoleVLA: Lightweight Vision-Language-Action Model with Deep State Space Models for Mobile Manipulation
- Symphony: A Cognitively-Inspired Multi-Agent System for Long-Video Understanding
- Do Understanding and Generation Fight? A Diagnostic Study of DPO for Unified Multimodal Models
- Multimodal Connectome Fusion via Cross-Attention for Autism Spectrum Disorder Classification Using Graph Learning
📚 RAG & Memory
118篇
该研究方向持续发展，产出多篇高质量论文。
- The Phasor Transformer: Resolving Attention Bottlenecks on the Unit Circle
- Symphony: A Cognitively-Inspired Multi-Agent System for Long-Video Understanding
- Graph-Native Cognitive Memory for AI Agents: Formal Belief Revision Semantics for Versioned Memory Architectures
- OPERA: Online Data Pruning for Efficient Retrieval Model Adaptation
- Token Coherence: Adapting MESI Cache Protocols to Minimize Synchronization Overhead in Multi-Agent LLM Systems
📁 Other
108篇
该研究方向持续发展，产出多篇高质量论文。
- Functorial Neural Architectures from Higher Inductive Types
- Interact3D: Compositional 3D Generation of Interactive Objects
- POaaS: Minimal-Edit Prompt Optimization as a Service to Lift Accuracy and Cut Hallucinations on On-Device sLLMs
- FederatedFactory: Generative One-Shot Learning for Extremely Non-IID Distributed Scenarios
- Prompt Engineering for Scale Development in Generative Psychometrics
🦾 Robotics & Embodied
61篇
该研究方向持续发展，产出多篇高质量论文。
- Proactive Rejection and Grounded Execution: A Dual-Stage Intent Analysis Paradigm for Safe and Efficient AIoT Smart Homes
- AnoleVLA: Lightweight Vision-Language-Action Model with Deep State Space Models for Mobile Manipulation
- Security Assessment and Mitigation Strategies for Large Language Models: A Comprehensive Defensive Framework
- SCAN: Sparse Circuit Anchor Interpretable Neuron for Lifelong Knowledge Editing
- Resilience Meets Autonomy: Governing Embodied AI in Critical Infrastructure
🏥 Healthcare
52篇
该研究方向持续发展，产出多篇高质量论文。
- Sample-Efficient Adaptation of Drug-Response Models to Patient Tumors under Strong Biological Domain Shift
- Empowering Chemical Structures with Biological Insights for Scalable Phenotypic Virtual Screening
- Pathology-Aware Multi-View Contrastive Learning for Patient-Independent ECG Reconstruction
- Security Assessment and Mitigation Strategies for Large Language Models: A Comprehensive Defensive Framework
- Federated Learning for Privacy-Preserving Medical AI
💻 Code
43篇
该研究方向持续发展，产出多篇高质量论文。
- SWE-QA-Pro: A Representative Benchmark and Scalable Training Recipe for Repository-Level Code Understanding
- Empowering Chemical Structures with Biological Insights for Scalable Phenotypic Virtual Screening
- Intent Formalization: A Grand Challenge for Reliable Coding in the Age of AI Agents
- SWE-Skills-Bench: Do Agent Skills Actually Help in Real-World Software Engineering?
- Exploring different approaches to customize language models for domain-specific text-to-code generation
📝 NLP
41篇
该研究方向持续发展，产出多篇高质量论文。
- AnoleVLA: Lightweight Vision-Language-Action Model with Deep State Space Models for Mobile Manipulation
- Intent Formalization: A Grand Challenge for Reliable Coding in the Age of AI Agents
- Exploring different approaches to customize language models for domain-specific text-to-code generation
- Prompt Programming for Cultural Bias and Alignment of Large Language Models
- MVHOI: Bridge Multi-view Condition to Complex Human-Object Interaction Video Reenactment via 3D Foundation Model
🔥 本周亮点论文
TrinityGuard: A Unified Framework for Safeguarding Multi-Agent Systems
With the rapid development of LLM-based multi-agent systems (MAS), their significant safety and security concerns have emerged, which introduce novel risks going beyond single agents or LLMs. Despite attempts to address these issues, the existing literature lacks a cohesive safeguarding system specialized for MAS risks. In this work, we introduce TrinityGuard, a comprehensive safety evaluation and monitoring framework for LLM-based MAS, grounded in the OWASP standards. Specifically, TrinityGuard...
TrinityGuard: A Unified Framework for Safeguarding Multi-Agent Systems
With the rapid development of LLM-based multi-agent systems (MAS), their significant safety and security concerns have emerged, which introduce novel risks going beyond single agents or LLMs. Despite attempts to address these issues, the existing literature lacks a cohesive safeguarding system specialized for MAS risks. In this work, we introduce TrinityGuard, a comprehensive safety evaluation and monitoring framework for LLM-based MAS, grounded in the OWASP standards. Specifically, TrinityGuard...
Harm or Humor: A Multimodal, Multilingual Benchmark for Overt and Covert Harmful Humor
Dark humor often relies on subtle cultural nuances and implicit cues that require contextual reasoning to interpret, posing safety challenges that current static benchmarks fail to capture. To address this, we introduce a novel multimodal, multilingual benchmark for detecting and understanding harmful and offensive humor. Our manually curated dataset comprises 3,000 texts and 6,000 images in English and Arabic, alongside 1,200 videos that span English, Arabic, and language-independent (universal...
Harm or Humor: A Multimodal, Multilingual Benchmark for Overt and Covert Harmful Humor
Dark humor often relies on subtle cultural nuances and implicit cues that require contextual reasoning to interpret, posing safety challenges that current static benchmarks fail to capture. To address this, we introduce a novel multimodal, multilingual benchmark for detecting and understanding harmful and offensive humor. Our manually curated dataset comprises 3,000 texts and 6,000 images in English and Arabic, alongside 1,200 videos that span English, Arabic, and language-independent (universal...
UniSAFE: A Comprehensive Benchmark for Safety Evaluation of Unified Multimodal Models
Unified Multimodal Models (UMMs) offer powerful cross-modality capabilities but introduce new safety risks not observed in single-task models. Despite their emergence, existing safety benchmarks remain fragmented across tasks and modalities, limiting the comprehensive evaluation of complex system-level vulnerabilities. To address this gap, we introduce UniSAFE, the first comprehensive benchmark for system-level safety evaluation of UMMs across 7 I/O modality combinations, spanning conventional t...
Directional Embedding Smoothing for Robust Vision Language Models
The safety and reliability of vision-language models (VLMs) are a crucial part of deploying trustworthy agentic AI systems. However, VLMs remain vulnerable to jailbreaking attacks that undermine their safety alignment to yield harmful outputs. In this work, we extend the Randomized Embedding Smoothing and Token Aggregation (RESTA) defense to VLMs and evaluate its performance against the JailBreakV-28K benchmark of multi-modal jailbreaking attacks. We find that RESTA is effective in reducing atta...
MultihopSpatial: Multi-hop Compositional Spatial Reasoning Benchmark for Vision-Language Model
Spatial reasoning is foundational for Vision-Language Models (VLMs), particularly when deployed as Vision-Language-Action (VLA) agents in physical environments. However, existing benchmarks predominantly focus on elementary, single-hop relations, neglecting the multi-hop compositional reasoning and precise visual grounding essential for real-world scenarios. To address this, we introduce MultihopSpatial, offering three key contributions: (1) A comprehensive benchmark designed for multi-hop and c...
SWE-Skills-Bench: Do Agent Skills Actually Help in Real-World Software Engineering?
Agent skills, structured procedural knowledge packages injected at inference time, are increasingly used to augment LLM agents on software engineering tasks. However, their real utility in end-to-end development settings remains unclear. We present SWE-Skills-Bench, the first requirement-driven benchmark that isolates the marginal utility of agent skills in real-world software engineering (SWE). It pairs 49 public SWE skills with authentic GitHub repositories pinned at fixed commits and requirem...
👥 作者合作网络
节点大小表示论文数量，连线表示合作关系
💡 技术创新趋势
🤖 智能体架构演进
多智能体协作、工具链整合、长期记忆机制成为研究热点，推动 AI Agent 向更复杂任务场景拓展。
🧠 推理能力提升
思维链优化、多步推理增强、逻辑演绎能力持续改进，大模型复杂问题求解能力显著提升。
🛡️ 安全对齐研究
模型安全、对抗防御、价值对齐研究持续深化，确保 AI 系统可靠可控成为关键议题。
⚡ 效率优化突破
量化压缩、推理加速、轻量化部署技术持续进步，推动大模型在边缘设备落地应用。
评论