You can login if you already have an account or register by clicking the button below.
Registering is free and all you need is a username and password. We never ask you for your e-mail.
[+]Frenchgeek0 points1 point1 point
ago
(edited ago)
[–]Frenchgeek0 points
1 point
1 point
(+1|-0)
ago
(edited ago)
To note, the Go AI in question is an hybrid using a classic tree-searching algorithm with a neural network taking the place of an evaluation function.
Because we have no idea what a good evaluation function for Go look like.
( as for why the classic tree search : it turn out neural networks simply can't do it effectively. Also finding practical uses to games trained neural networks seem to be pretty hard, since we don't know how or why they work. )
( https://medium.com/backchannel/has-deepmind-really-passed-go-adc85e256bec )
view the rest of the comments →
[–] Frenchgeek 0 points 1 point 1 point (+1|-0) ago (edited ago)
To note, the Go AI in question is an hybrid using a classic tree-searching algorithm with a neural network taking the place of an evaluation function.
Because we have no idea what a good evaluation function for Go look like.
( as for why the classic tree search : it turn out neural networks simply can't do it effectively. Also finding practical uses to games trained neural networks seem to be pretty hard, since we don't know how or why they work. )
( https://medium.com/backchannel/has-deepmind-really-passed-go-adc85e256bec )