DeepMind have rocked the world of reinforcement learning over the last three years with their AlphaZero AI. Following the incredible achievement of beating Grandmaster Lee Sedol at the game of Go (described as decades ahead of its time), they started again with a “Zero” approach to learning from first principles with no human input. This new AI learned how to play chess in seven hours, beating the world’s greatest chess computer program, Stockfish, and is now learning to play StarCraft at a professional level.
This session aims to give an introduction to the principles of reinforcement learning, and the techniques that AlphaZero uses to achieve its superhuman performance.
The objective of the talk is to give a gentle introduction to the field of reinforcement learning and how, starting from random play, you can train an expert level AI in a complex game. Common pitfalls are discussed and how techniques used by the AlphaZero project can be used to work around these.
No background knowledge necessary