Previous versions of AlphaGo initially trained on thousands of human amateur and professional games to learn how to play Go. AlphaGo Zero skips this step and learns to play simply by playing games against itself, starting from completely random play. In doing so, it quickly surpassed human level of play and defeated the previously published champion-defeating version of AlphaGo by 100 games to 0.Pretty impressive stuff. DeepMind hopes similar techniques can be used to train AI agents to solve challenging real-world problems, like protein folding, cutting energy consumption, or material science issues.
Google AlphaGo Zero teaches itself how to beat world champions (video)
Posted on Friday, October 20 2017 @ 10:36 CEST by Thomas De Maesschalck
Here's a new video from Google's DeepMind unit. It talks about how the AlphaGo Zero algorithm trained itself to master the ancient Chinese game of Go. The computer required no human input and rapidly evolved into world's best Go player. DeepMind says it took just 70 hours for AlphaGo Zero to develop superhuman-like capabilities in this game.