|
BGonline.org Forums
AI wins final match won series 4-1
Posted By: Maik Stiebler In Response To: AI wins final match won series 4-1 (AP)
Date: Wednesday, 16 March 2016, at 7:17 p.m.
I guess the breakthrough is just that huge jump in playing strength. As a side note, machine learning was an important part of it. Are you saying that no success of machine learning can ever be remarkable anymore, just because it worked in backgammon 20 years ago?
The use of neural networks and machine learning is not a novelty even in Go programming. For example, it has been tried in WinHonte, a program by Fredrik Dahl, creator of Jellyfish. I don't know if and doubt that before AlphaGo, there ever was a program that fully relied on neural nets for the global board evaluation though. Actually, AlphaGo does not only use an NN for evaluation, but also another one that outputs a heat map of the 19x19 board indicating hot points to play a good next move to. The distinction between these two may sound weird (one could think it should be possible to construct such a heat map from the evaluation function), but it seems to be efficient to use the latter net as a move generator for their version of the Monte Carlo tree search.
|
BGonline.org Forums is maintained by Stick with WebBBS 5.12.