Connect Four AI (AlphaGo Zero style)
Reinforcement Learning allows the computer to play games against itself to gradually improve its play following reward-based learning algorithms without any major external influences or input.
This technique was made famous when the computer program AlphaGo Zero, with only 40 days of training via pure self-play, was able to beat the previous version of the AI AlphaGo which had already beaten South Korean Go professionals like Lee Sedol by an impressive score of 4:1. Later versions like Alpha Zero was able to achieve superhuman capabilities in chess/shogi/go with only 24 hours of training!
Click here to see the github repository and run it yourself!