DeepMinds AlphaStar AI vs Team Liquid’s Grzegorz "MaNa" Komincz 20190124

Описание: DeepMind developed AlphaGo and AlphaZero, both successful deep convoluted neural networks, learning to win at boardgames such as go and chess, learning to win by selfplay. These are great steps forward for AI research.

Now they have turned their attention to StarCraft II. The results are impressive, but still leave room for improvement. So far only one match has been released where the AI is using the new Camera Interface, something that clearly needed to be implemented to level the playing field. Let me elaborate:

First they presented an AI using a RAW interface which, in hindsight, was winning mostly due to higher burst APM used as micro during battles. Now they have implemented the camera interface, something that is much more like how humans can play SC2. Only one game has so far been released and even in this game the AI is using a very high APM bursts, but maybe not higher than what humans might be able to reach.

For more information:
https://deepmind.com/research/alphast...

Video is just the replay from AlphaStars point of view with APM shown. For comments, I suggest you check out other channels:
https://www.youtube.com/results?searc...

Games using the raw interface:
   • DeepMind StarCraft II Demonstration  

WinterStarCraft analysis of all the games, starting at this game:
   • StarCraft 2 Pros vs Google's Deepmind AI (...