视频选题榜 2026-08-01

• https://arxiv.org/abs/2607.27205 · TurboVLA: Real-Time Vision-Language-Action Model at 32 Hz on an RTX 4090 with <1 GB VRAM · TurboVLA 把 VLA 从 LLM-centric V→L→A 改写为 V+L→A 直接融合(DINOv3 + BERT 独立编码 + 双向轻量交互 + 紧凑 decoder)· 0.2B / 31.2 ms / 0.9 GB / 32 Hz / LIBERO 97.7% · round=R1- https://arxiv.org/abs/2607.26520 · A Graph-Native Bitemporal Memory Store for Conversational AI Agents · Agent-local Neo4j 属性图 + HNSW + 双时态数据模型(60 题抽样 R@10 46.7% / 80% / 37.5%)· round=R2
• https://arxiv.org/abs/2607.26637 · Filesystem-Based Memory for LLM Agents: Organization, Evolution, and Sustainability · 文件系统作为 LLM Agent 长期记忆默认媒介 · 5 RQ + 50/100/200 session horizon + Management/Search/Execution 三角色 + 工具集 = 杠杆 · Sizhe Zhou 等 11 作者 · round=R3