Hacker Read

espadrine | karma 7290 | avg karma 6.06 · 2017-12-07 17:45:20

> AlphaZero won some games in a romantic style

The difference in style is likely influenced by the insertion of historical boards as input to the neural network.

The sequence of moves are therefore more likely to look related to one another.

grondilu | karma 1865 | avg karma 1.8 · 2017-12-07 17:48:04

The neural network was only fed with games from self-play. No historical game whatsoever was given. At least that's Deepmind's claim.

espadrine | karma 7290 | avg karma 6.06 · 2017-12-08 10:39:46

I was misunderstood. By historical, I mean, it includes the past N boards, which pushes the network to make actions correlated to the previous actions performed.

That is similar to how humans play.

reply