In our previous work we demonstrated how to achieve superhuman performance in the game of Go, a long standing challenge for AI. Subsequently we also showed how to achieve the same level of performance tabula rasa, without builtin human knowledge or hand engineered features.
In our most recent work we further generalise our algorithm: AlphaZero exceeds the state of the art in Go, Chess as well as Shogi without game specific adaptions beyond the rules.
Presenter: Julian Schrittwieser is a senior software engineer on DeepMind’s AlphaGo team, second author on the recent AlphaGo Zero paper and authored the company’s first landmark paper on AlphaGo at the age of 23. He is the core developer of AlphaGo Zero, working on everything from the main search algorithm, the training framework to support for new hardware, while also creating network architectures to tackle more complex games and real-world science problems.