Mastering the Game of Go with Deep Neural Networks and Tree Search .

Abstract

AlphaGo defeated the world champion Lee Sedol in Go, a game previously thought to be decades away from computer mastery.

Techniques

• Policy Network: Predicts expert moves
• Value Network: Evaluates board positions
• Monte Carlo Tree Search: Planning algorithm
• Self-Play: Training by playing against itself

Impact

Demonstrated that deep RL could solve complex strategic problems. Inspired AlphaZero (chess, shogi) and AlphaFold (protein folding).