Learning From Scratch with AlphaGo

---

A team from Google led by Hassabis (2017) has recently published a paper in Nature titled Mastering the game of Go without human knowledge. Their motivation is the development of an algorithm that learns from scratch - without human input.

As per the researchers:

"The paper introduces AlphaGo Zero, the latest evolution of AlphaGo, the first computer program to defeat a world champion at the ancient Chinese game of Go. Zero is even more powerful and is arguably the strongest Go player in history." [source]

Earlier versions of AlphaGo were trained using thousands of human amateur and professional games. However, AlphaGo Zero is different because it trains from scratch by playing against itself - with a random initialization.

Here's the interesting part:

AlphaGo Zero quickly surpasses human level and when playing against earlier versions of AlphaGo, it performed 100 to 0. That's pretty amazing, if you ask me. The type of learning that powers Zero's performance is a form of reinforcement learning:

"The system starts off with a neural network that knows nothing about the game of Go. It then plays games against itself, by combining this neural network with a powerful search algorithm. As it plays, the neural network is tuned and updated to predict moves, as well as the eventual winner of the games." [source]