中央研究院 資訊科學研究所

In this talk, I will present the research on the development of CGI, an AlphaGo/AlphaZero-based Computer Go Program. First, I will briefly summarize our research achievements including some awards and honors. Then, I will present three of the past research topics. (a) Proposed a new approach to a novel value network architecture for the game of Go. This is the first AlphaGo-like program that can play handicap games. (b) Applied Population Based Training (PBT) to training a Go program CGI. Based on this method, CGI obtains a winrate of 74% against ELF OpenGo (SOTA), developed by Facebook. (c) Proposed a novel approach to strength adjustment for MCTS-based game-playing programs. Based on this method, CGI can cover strengths ranging from beginners to super-human players for Go. Finally, I will also describe the ongoing research topics.

Ti-Rong Wu is currently a postdoctoral fellow in the Department of Computer Science at National Yang Ming Chiao Tung University. He received his Ph.D. degree in Computer Science from National Chiao Tung University in 2020. His research interests include computer games, reinforcement learning, and deep reinforcement learning. He led a group in the CGI lab that developed a computer Go program, named CGI Go Intelligence, which won many competitions, particularly the second place in the First World AI Go Open in 2017, in which defeated FineArt, developed by Tencent. The research results have also been published in several top-tier journals and conferences.