PDF] MONTE CARLO TREE SEARCH: A TUTORIAL

Por um escritor misterioso

Last updated 22 dezembro 2024

PDF] MONTE CARLO TREE SEARCH: A TUTORIAL

This tutorial provides an introduction to MCTS, including a review of its history and relationship to a more general simulation-based algorithm for Markov decision processes (MDPs), a demonstration of the basic mechanics of the algorithms via decision trees and the game of tic-tac-toe; and its use in AlphaGo and AlphaZero. Monte Carlo tree search (MCTS) is a general approach to solving game problems, playing a central role in Google DeepMind’s AlphaZero and its predecessor AlphaGo, which famously defeated the (human) world Go champion Lee Sedol in 2016 and world #1 Go player Ke Jie in 2017. Starting from scratch without using any domain-specific knowledge (other than the game rules), AlphaZero defeated not only its ancestors in Go but also the best computer programs in chess (Stockfish) and shogi (Elmo). In this tutorial, we provide an introduction to MCTS, including a review of its history and relationship to a more general simulation-based algorithm for Markov decision processes (MDPs) published in a 2005 Operations Research article; a demonstration of the basic mechanics of the algorithms via decision trees and the game of tic-tac-toe; and its use in AlphaGo and AlphaZero.