4 ms·
"They trained the system for five hypothetical steps and a million mini-batches (i.e., small batches of training data) of size 2,048 in board games and size 1,0
by smithmayowa 7y ago
"They trained the system for five hypothetical steps and a million mini-batches (i.e., small batches of training data) of size 2,048 in board games and size 1,024 in Atari, which amounted to 800 simulations per move for each search in Go, chess, and shogi and 50 simulations for each search in Atari"
Because of this I am presuming.