Convolutional Monte Carlo Rollouts in Go

TL;DR

This paper introduces a Monte Carlo Tree Search approach for Go that integrates convolutional neural networks throughout, utilizing batch processing and GPU acceleration to improve performance against existing programs.

Contribution

It presents a novel MCTS-based Go program that employs convolutional networks in all components, with batch processing and GPU-accelerated rollouts for enhanced efficiency.

Findings

Achieves strong win rates against open source Go programs

Attains competitive results against state-of-the-art convolutional net-based Go programs

Demonstrates the effectiveness of convolutional networks in all parts of MCTS for Go

Abstract

In this work, we present a MCTS-based Go-playing program which uses convolutional networks in all parts. Our method performs MCTS in batches, explores the Monte Carlo search tree using Thompson sampling and a convolutional network, and evaluates convnet-based rollouts on the GPU. We achieve strong win rates against open source Go programs and attain competitive results against state of the art convolutional net-based Go-playing programs.