Mastering the game of Go with deep neural…

This paper asks how a computer can choose a strong Go move when the game tree is far too large to enumerate. Read it to see how a learned policy concentrates search on promising moves, how a value network estimates who is ahead before a game ends, and how Monte Carlo tree search turns those estimates into professional-level play under a large compute budget.

Reading focus: How a policy network reduces the effective breadth of search while a value network reduces the effective depth. Why expert imitation, self-play reinforcement learning, and independent value-training positions solve different parts of the problem. How policy priors, visit counts, value estimates, and rollouts combine inside Monte Carlo tree search, and which compute and evaluation limits qualify the result.

Nature 2016. Silver et al.. 35 min read, easy difficulty.