Comments on: What’s up with go? http://www.vetta.org/2009/04/whats-up-with-go/ Thu, 09 Sep 2010 11:22:32 +0000 hourly 1 http://wordpress.org/?v=3.0.1 By: Denis http://www.vetta.org/2009/04/whats-up-with-go/comment-page-1/#comment-19421 Denis Thu, 23 Apr 2009 18:34:35 +0000 http://www.vetta.org/?p=288#comment-19421 I suggest to you this this interesting link http://k21st.wordpress.com/2009/03/04/next-big-future-ai-milestone-supercomputer-given-67-stone-handicap-able-to-win-professional-19x19-go-games/ I suggest to you this this interesting link

http://k21st.wordpress.com/2009/03/04/next-big-future-ai-milestone-supercomputer-given-67-stone-handicap-able-to-win-professional-19×19-go-games/

]]> By: Shane Legg http://www.vetta.org/2009/04/whats-up-with-go/comment-page-1/#comment-19413 Shane Legg Wed, 22 Apr 2009 10:11:20 +0000 http://www.vetta.org/?p=288#comment-19413 A naive application of elementary RL methods fail with go due to the branching factor, as you suspected. There are however people trying RL approaches that use the MC Tree Search as part of the algorithm. I haven't read about this research myself, I've just been told about it. I believe it's still actively being researched. A naive application of elementary RL methods fail with go due to the branching factor, as you suspected. There are however people trying RL approaches that use the MC Tree Search as part of the algorithm. I haven’t read about this research myself, I’ve just been told about it. I believe it’s still actively being researched.

]]> By: Seinberg http://www.vetta.org/2009/04/whats-up-with-go/comment-page-1/#comment-19409 Seinberg Wed, 22 Apr 2009 00:15:03 +0000 http://www.vetta.org/?p=288#comment-19409 This is a naive comment without Googling first, but has there been much research into reinforcement learning with Go? It's been successful in Backgammon, but of course Go (and even Chess) are significantly more complicated and have much higher branching factors so maybe that would create policies that are just as computationally expensive as brute force techniques. This is a naive comment without Googling first, but has there been much research into reinforcement learning with Go? It’s been successful in Backgammon, but of course Go (and even Chess) are significantly more complicated and have much higher branching factors so maybe that would create policies that are just as computationally expensive as brute force techniques.

]]>