jump to main area
:::
A- A A+

Seminars

The Techniques from Erica Winning Gold to AlphaGo Beating the World Go Champion 從Erica奪金到AlphaGo打敗棋王之技術

  • 2016-09-05 (Mon.), 10:30 AM
  • Recreation Hall, 2F, Institute of Statistical Science
  • Prof. Shun-Shii Lin
  • Department of Computer Science and Information Engineering, National Taiwan University

Abstract

It was an exciting moment when AlphaGo beat the world Go champion, which was also a historical moment for the advancement of AI. Hence, I would like to speak on the techniques used from Erica’s winning a gold medal to AlphaGo’s beating Lee Sedol. ????? Ko plays a very important role in Go, but most early-developed computer Go programs still could not handle Ko fights by 2001.? Then, Aja Huang and I started to tackle this difficult problem at NTNU. Utilizing the principle of game theory, we obtained the best strategies for the Go programs to gain maximum or loss minimum profit when dealt with Ko fights. ???? ?In 2011, Aja Huang, R?mi Coulom and I proposed some new heuristics of Monte Carlo Tree Search focusing on the application of Simulation Balancing and various time management schemes for Go. These results led our Go program ERICA to win the gold medal in the Go tournament at the 2010 Computer Olympiad, Japan. To date, many techniques of ERICA, like Monte Carlo Tree Search, were still used in AlphaGo. Furthermore, deep learning techniques were successfully applied in AlphaGo to boost its strength. ????? Due to these outstanding results, a road incidentally built at NTNU during the past 10 years advanced the birth of AlphaGo which beat the world Go champion Lee in 2016.

Update:2024-12-02 21:15
scroll to top