[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

XG Update (Crossposted from BG News FB group)

Posted By: Timothy Chow
Date: Friday, 14 July 2023, at 1:56 a.m.

In Response To: XG Update (Crossposted from BG News FB group) (Frank Berger)

I totally agree with your fairy dust comment. In the years since AlphaZero's debut, it has become clear that there's a lot of tweaking that needs to be done to achieve good performance. See for example this presentation about the difficulties of reproducing AlphaZero. See also this article about beating top Go programs by exploiting surprising weaknesses that they have.

But I don't agree with the attribution of the success of AlphaZero to the difference between 1D and 2D games. I accept that using a convolutional neural net may not be as much of a win in backgammon as in Go. But convolution is far from the only trick in the deep learning toolbox nowadays. The simple fact is that we now have a much better understanding of how to take advantage of multiple neural network layers, and huge amounts of processing power. There was never any reason in principle that a neural network couldn't learn how to roll a prime all the way around the board, but in practice, the traditional machine learning methods and traditional amounts of computing power would come up short. I expect nowadays it should be possible to overcome those obstacles and build a net that can (for example) handle the positions in Wachtel's book easily, not to mention even crazier positions.

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.