There were many advances in Deep Learning and AI in 2017, but few generated as much publicity and interest as DeepMind’s AlphaGo Zero. This program was truly a shocking breakthrough: not only did it beat the prior version of AlphaGo — the program that beat 17 time world champion Lee Sedol just a year and a half earlier — 100–0, it was trained without any data from real human games.

Xavier Amatrain called it “more [significant] than anything…in the last 5 years” in Machine Learning. Both AlphaGo and AlphaGo Zero evaluated the Go board and chose moves using a combination of two methods: AlphaGo and AlphaGo Zero both worked by cleverly combining these two methods.

Let’s look at each one in turn: Go is a sufficiently complex game that computers can’t simply search all possible moves using a brute force approach to find the best one (indeed, they can’t even come close). The best Go programs prior to AlphaGo overcame this by using “Monte Carlo Tree Search” or MCTS.

At a high level, this method involves initially exploring many possible moves on the board, and then focusing this exploration over time as certain moves are found to be more likely to lead to wins than others. Both AlphaGo and AlphaGo Zero use a relatively straightforward version of MCTS for their “lookahead”, simply using many of the best practices listed in the Monte Carlo Tree Search Wikipedia page to properly manage the tradeoff between exploring new sequences of move or more deeply explore already-explored sequences (for more, see the details in the “Search” section under “Methods” in the original AlphaGo Paper published in Nature).

Though, MCTS had been the core of all successful Go programs prior to AlphaGo, it was DeepMind’s clever combination of this technique with a neural network-based “intuition” that allowed it to surpass human performance. DeepMind’s major innovation with AlphaGo was to use deep neural networks to understand the state of the game, and then use this understanding to intelligently guide the search of the MCTS. Read more from…

thumbnail courtesy of