MatPlus.Net

 Website founded by
Milan Velimirović
in 2006

21:08 UTC
ISC 2021
 
  Forum*
 
 
 

Username:

Password:

Remember me

 
Forgot your
password?
Click here!
SIGN IN
to create your account if you don't already have one.
CHESS
SOLVING

Tournaments
Rating lists
1-Oct-2021

B P C F





 
 
MatPlus.Net Forum General AlphaFold
 
You can only view this page!
(1) Posted by Neal Turner [Tuesday, Dec 1, 2020 16:02]

AlphaFold


So the DeepMind people have cracked one of the toughest scientific puzzles - protein folding - with their AlphaFold system.
Of course the the earlier systems they made to master Go & Chess were just 'practice', used to develop and evolve their ideas in a constrained, well-defined environment.

Isn't it good that we're so low-profile and off the radar!
I'm guessing it wouldn't be too much for them to set up their system to compose chess problems.
Then we could all go back to doing something useful - or not - I suspect most of us are too far gone!
 
(Read Only)pid=20152
(2) Posted by Frank Richter [Tuesday, Dec 1, 2020 17:28]

It would be a very interesting experiment to set up their system to compose chess problems.
"Simply" (?) enter a lot of database records from PDB, WinChloe and other problem databases, add some formal clarifications and rules ... and start processing.
What do you think, would the AlphaComposer be able to create really new, original and interesting chess problems or not?
 
(Read Only)pid=20153
(3) Posted by Rewan Demontay (Real Name: James Malcom) [Tuesday, Dec 1, 2020 17:32]

We could use it to do something useful, such as programming it to find the so called "Perfect Babson," by giving it all known full and 3/4 Babsons as per Frank's above comment.
 
 
(Read Only)pid=20154
(4) Posted by Hauke Reddmann [Tuesday, Dec 1, 2020 17:55]

As a computer science student (even with some
experience in neural networks) I say: No chance.

The problem is the cost function. Even the
landscape of protein folding looks orderly
when compared to problem chess. Cf. Azlan Iqbal
and his bold attempts (with even some more
explicite built-in knowledge). No cigar so far.

I predict the reverse methodology would be faster
and more promising:
a) Define some formal theme
b) Mine a load of positions
c) Profit!
I might even suspect the one or other problemist
if I wouldn't fear a libel suit :-)

Hauke
 
 
(Read Only)pid=20155

No more posts


MatPlus.Net Forum General AlphaFold