Re: [math-fun] Deep learning v deep positions
I didn't see the start of this thread so don't know if it's just about Go or just about Alpha<x>. However, we have the advent of chess engines which rely more (than traditionally) on intelligent rather than deep search. Neural networks are challenging traditional 'alpha-beta cut-off' search, not just with AlphaZero but with LeelaChess and one other engine, making its first appearance in TCEC, Season 13. TCEC13, http://tcec.chessdom.com/live.php , should have kicked off on or before August 3rd according to my latest advice. The two engines above will be using 2x GTX 1080ti GPUs but not Google's TPUs. The improving LeelaChess may arguably be 'better than human' at chess but has not yet been competitive with TCEC engines in its lowest division. g
participants (1)
-
Guy Haworth