Neural networks (NN) explained

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
zenpawn
Posts: 299
Joined: Sat Aug 06, 2016 6:31 pm
Location: United States

Neural networks (NN) explained

Post by zenpawn » Wed Dec 20, 2017 3:06 pm

This video (series) provides a beautiful introduction to neural networks:

https://www.youtube.com/watch?v=aircAru ... x_ZCJB-3pi

Daniel Anulliero
Posts: 688
Joined: Fri Jan 04, 2013 3:55 pm
Location: Nice

Re: Neural networks (NN) explained

Post by Daniel Anulliero » Wed Dec 20, 2017 3:10 pm

Thanks for the link!
And soon , RMzero :wink:
Cool for A V D :lol:

Henk
Posts: 5833
Joined: Mon May 27, 2013 8:31 am

Re: Neural networks (NN) explained

Post by Henk » Wed Dec 20, 2017 3:57 pm

If you want to follow the Alpha zero approach you can only reuse the move generator and the user interface. Search need to be replaced and evaluation code is not used. Maybe board representation need to be optimized too. Perhaps using only bit boards is good enough.

A lot of work. Also complicated stuff. Hi hi hi . I'm getting mad. The time I'm finished they have invented something much and much better.

But isn't that with all software. It only lasts a few years and then it's totally out of date.

CheckersGuy
Posts: 273
Joined: Wed Aug 24, 2016 7:49 pm

Re: Neural networks (NN) explained

Post by CheckersGuy » Wed Dec 20, 2017 4:06 pm

True. I think the speed of the moveGenerator and the underlying board representation isn't that important when you are "bottlenecked" by the NN (in terms of speed) :D

Henk
Posts: 5833
Joined: Mon May 27, 2013 8:31 am

Re: Neural networks (NN) explained

Post by Henk » Wed Dec 20, 2017 4:13 pm

My move generator also sorts the moves. But that makes no sense anymore.


Evaluating a connection in a neural network is most expensive operation. Can't do much about that to speed it up. Maybe removing redundant nodes. Weight decay.

pilgrimdan
Posts: 402
Joined: Sat Jul 02, 2011 8:49 pm

Re: Neural networks (NN) explained

Post by pilgrimdan » Thu Dec 21, 2017 8:55 am

zenpawn wrote:This video (series) provides a beautiful introduction to neural networks:

https://www.youtube.com/watch?v=aircAru ... x_ZCJB-3pi
thanks ...

TommyTC
Posts: 18
Joined: Thu Mar 30, 2017 6:52 am

Re: Neural networks (NN) explained

Post by TommyTC » Thu Dec 21, 2017 1:26 pm

zenpawn wrote:This video (series) provides a beautiful introduction to neural networks:

https://www.youtube.com/watch?v=aircAru ... x_ZCJB-3pi
Thank you!
I know a bit about computers and a bit about chess, and have dabbled a little with computer chess engines.

With all this talk of AlphaZero and viewing a lot of video and comments about AlphaZero, I never saw anything that tried to describe neural networks. After viewing this video, I now have an idea of what a neural network is!

Post Reply