To educate their code, the researchers chose three YouTube gameplay videos for each of the three titles: Montezuma’s Revenge, Pitfall, and Private Eye. Each game had its own agent, which had to map the actions and features of the title into a form it could understand. The team used two methods: temporal distance classification (TDC), and cross-modal temporal distance classification (CDC).More at The Register. It will be interesting to see if these deep learning systems will soon be able to learn more complex games.
DeepMind teaches AI to play games by watching YouTube footage
Posted on Friday, June 01 2018 @ 12:06 CEST by Thomas De Maesschalck
Google's DeepMind unit has created a new deep learning system that learns itself how to play video game classics by watching YouTube footage to analyze how humans tackle the games.