3 ms·

Also (3) training a dual policy & value network that can benefit from a single shared representation of the game