Recent Posts

Heads of Security: Learned Attention Gates for Robust Prompt Injection Defense

Heads of Security：基于可学习注意力门控的提示词注入主动防御

10/29/2026

arxivaics.lg

ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning for Robotics

ThinkWorld：具身世界想象与动力链推理强化的通用机器人操作

10/29/2026

arxivaics.ro

Look Only When Needed: Certificate-Guided Visual Routing for Efficient Diffusion VLMs

LOWN：按需观照，基于一致性认证的扩散视觉语言模型高效视觉路由

10/28/2026

arxivaics.cv

SWIM: Learning What Matters from Behaviors for JEPA World Models

SWIM：因行知重，基于逆动力学行为加权的 JEPA 具身世界模型

10/28/2026

arxivaics.ro

Auto-Research Arena: Benchmarking and Advancing Autonomous AI Research Systems

Auto-Research Arena：自主科研智能体评测与突破的通用基准

10/27/2026

arxivaics.lg

DecisionRAG: Retriever-Free Retrieval via Hierarchical Decision Models

DecisionRAG：告别检索器，基于分层决策模型的免检索式文档导引

10/27/2026

arxivaics.ai

If at First You Don't Succeed, Backtrack: Lightweight Recovery for Frozen VLAs

一次失手就倒带重来：基于运动回溯的冻结 VLA 具身故障恢复机制

10/26/2026

arxivaics.ro

J-Comm: Jacobian-Guided Latent Communication between LLM Agents

J-Comm：基于雅可比矩阵引导的跨智能体潜空间连续通信

10/26/2026

arxivaics.cl

UniAction: Unified Action Conditioning for World Models

UniAction：面向世界模型的跨模态统一动作条件化框架

10/26/2026

arxivaics.cv

Where Rounding Happens Matters: Rounding-Location Alignment Stabilizes LLM Reinforcement Learning

舍入发生的位置至关重要：对齐舍入位置彻底稳定大模型强化学习训练

10/26/2026

arxivaics.lg

Causal Chain of Thought: Mitigating Attention Leakage via Context Isolation

因果思维链 (CausalCoT)：基于上下文隔离阻断全向注意力泄漏

10/23/2026

arxivaics.ai

Direct Pass@k Optimization for Thought Pruning: Balancing Quality and Diversity in Test-Time Reasoning

基于直接 Pass@k 优化的思维前缀剪枝：统合测试期推理的质量与多样性

10/23/2026

arxivaics.ai