4 comments

[ 23.8 ms ] story [ 1441 ms ] thread
It would be interesting to know if there were similar techniques employed for AlphaStar and AlphaGo, with DeepMind's use of LSTMs, etc...
From my understanding AlphaStar used a LSTM to determine the actions of its agents, but it had a transformer to draw attention to a subset of the inputs
It would be nice to get an in-depth overview of the techniques used behind AlphaStar...
DeepMind reduced the problem to search using the two Deep Neural Networks. MCTS is an old technique but it clearly worked here