Maru version 8.1

Maru is a computer Go program developed using deep reinforcement learning from randomly generated game records. The deep learning model of Maru incorporates large-kernel depthwise convolutions and multi-head attention, enabling it to efficiently grasp the overall state of the board. Its reinforcement learning procedure is inspired by approaches used in Katago and Gumbel AlphaZero, allowing it to efficiently learn a wide range of patterns. This page shows the improvement in playing strength through reinforcement learning and the results of self-play matches.

Maru is a sibling program of the computer Shogi program Gokaku. Maru shares the same deep learning model architecture, search algorithm, and reinforcement learning methodology as Gokaku.

This page shows the changes in playing strength during the reinforcement learning process of the latest version, Maru version 8.1. The Elo rating is calculated by playing matches against 60 models of nearby generations. The Elo ratings of other baseline programs are calculated based on their match results against Maru (these are not objective indicators of each program’s actual strength).

The executable program and model files for Maru are available on GitHub. The progress of reinforcement learning for previous version, Maru version 8.0, can be found on here. If you have any questions, please contact Atsushi Takeda.