Reinforcement Learning with TensorFlow

AlphaGo Zero

The first generation of AlphaGo was able to beat the professional Go players. In October 2017, Google DeepMind published the paper (https://www.nature.com/articles/nature24270) on AlphaGo Zero in Nature. AlphaGo Zero is the latest version of AlphaGo. Earlier versions of AlphaGo learnt to play the game after being trained on thousands of human games varying from amateur to professional games. But the final version of AlphaGo, that is AlphaGo Zero has learnt everything from scratch, that is from the first basic principle neither using any human data nor any human intervention and was able to achieve the highest level of performance. Thus, AlphaGo Zero learns to play the game of Go by playing against itself. One of the biggest feats ...