what we don't know
how many machines (and what hardware) did DM train AGZ on ?
DM explains many details of AGZ architecture, but parameter tuning is critical in neural net training and performance
did AGZ use (perhaps to start) AG neural net parameter values ? if yes, is it fair to say that AGZ learns tabula rasa ?
2017 dec AlphaZero self-learns Go, chess, shogi (Japanese chess)
2017-12-06 AZ-Stockfish 28-72-0 but did Stockfish have same resources as AZ ?
DeepMind story: right place, right time
Demi Hassabis, David Silver, Aja Huang, …
people + machines + MCTS + DCNN
CMPUT 496 follows CMPUT 396, more technical details
tutorial on game tree research