[Computer-go] Teaching Deep Convolutional Neural Networks to Play Go

[Hugh Perkins]

On 3/17/15, David Silver <davidstarsilver at gmail.com> wrote:
> Reinforcement learning is different to unsupervised learning. We used
> reinforcement learning to train the Atari games. Also we published a more
> recent paper (www.nature.com/articles/nature14236) that applied the same
> network to 50 different Atari games (achieving human level in around half).

Omg, the Atari paper is an awesome paper.  Cant believe I skipped over
it the first time.  I guess I was like "Oh, it's not Go, skip that one
for now :-)" :-D

It's really amazing, it's exactly what I was hoping someone could
achieve.  Well... "exactly"... I suppose "exactly" would mean, could
learn to play http://springrts.com :-)  Perhaps, "conceptually" is a
better word.  My idea was to just give the computer generic,
unlabelled, arrays as input, representing the map and stuff; and a set
of generic, unlabelled buttons as output, ie representing 'up' 'down',
etc.  But I had no idea how to train it :-)  And now, someone has come
up with a way to train such a device :-)