Xin Ding

Information Gain-based Rollout Policy Optimization: An Adaptive Tree-Structured Rollout Approach for Multi-Turn LLM Agents

Yijun Zhang, Fan Xu, Jiaxin Ding, et al.

Em-Garde: A Propose-Match Framework for Proactive Streaming Video Understanding

Yikai Zheng, Xin Ding, Yifan Yang, et al.

Em-Garde is a new system for understanding video streams in real-time that can proactively answer user questions about what's happening in videos. Instead of checking every frame to decide when to respond (which is slow and inaccurate), it converts user questions into visual search patterns and efficiently matches them against the incoming video stream, achieving better accuracy with less computational effort.