from Hacker News

Much More Efficient Self-Play Learning in Go

by jedharris on 6/14/20, 10:14 PM with 2 comments

• by jedharris on 6/14/20, 10:22 PM

  The author's most recent version of KataGo averages about 50 times greater learning efficiency from self-play.

  See <https://blog.janestreet.com/deep-learning-the-hardest-go-pro.... This covers developments since the initial post, plus the headline problem (which is still open).

• by jedharris on 6/14/20, 10:16 PM

  Related demos, repositories, papers and posts:

  More recent and extensive post: https://blog.janestreet.com/deep-learning-the-hardest-go-pro...

  Site to play against KataGo: https://online-go.com/player/592684/

  Arxiv paper on KataGo design: https://arxiv.org/abs/1902.10565

  GitHub repository of KataGo: https://github.com/lightvector/KataGo

  Another writer commenting on the history of Go AI, and significance of KataGo: https://brantondemoss.com/writing/kata.html