What algorithm does AlphaZero use

Here is an excellent summary of the architecture, by reddit author cbeak, > man, this is so simple and yet so powerful:Im danny rensch explains the alphazero match in a series of videos on twitch.It seems that so far, alphago, alphago zero and alphazero have primarily been used for solving games.By contrast, alphazero had not been taught.Make sure it gets better in the next loop.

In contrast to supervised learning where data is.In alphazero, rollouts are replaced by fetching predictions from the nn, and ucb1 is replaced by puct (polynomial upper confidence tree).(in a nod to the gaming mecca, mathematicians refer to approaches that involve some.They are two very different algorithms.Stockfish 8 had won a world computer chess championship in 2016.

The neural network for deepmind's alphazero is updated regularly.Alphazero could be used for a strategy game like chess or even shogi.Let the computer play against itself several games, recording the board at every move.Not only does it outperform all previous go players, human or machine, it does so after only three days of training time.