Affiliation(s):
Key Laboratory of Ethnic Language Intelligent Analysis and Security Governance of MOE, Minzu University of China, Beijing 100081, China;
moreAffiliation(s): Key Laboratory of Ethnic Language Intelligent Analysis and Security Governance of MOE, Minzu University of China, Beijing 100081, China; School of Information Engineering, Minzu University of China, Beijing 100081, China; Department of Advanced Manufacturing and Robotics, College of Engineering, Peking University, Beijing 100871, China;
less
Xiali LI, Yanyin ZHANG, Licheng WU, Yandong CHEN, Junzhi YU. TibetanGoTinyNet: a light weight U-Net style network for Zero learning of Tibetan Go[J]. Frontiers of Information Technology & Electronic Engineering,in press.https://doi.org/10.1631/FITEE.2300493
@article{title="TibetanGoTinyNet: a light weight U-Net style network for Zero learning of Tibetan Go", author="Xiali LI, Yanyin ZHANG, Licheng WU, Yandong CHEN, Junzhi YU", journal="Frontiers of Information Technology & Electronic Engineering", year="in press", publisher="Zhejiang University Press & Springer", doi="https://doi.org/10.1631/FITEE.2300493" }
%0 Journal Article %T TibetanGoTinyNet: a light weight U-Net style network for Zero learning of Tibetan Go %A Xiali LI %A Yanyin ZHANG %A Licheng WU %A Yandong CHEN %A Junzhi YU %J Frontiers of Information Technology & Electronic Engineering %P %@ 2095-9184 %D in press %I Zhejiang University Press & Springer doi="https://doi.org/10.1631/FITEE.2300493"
TY - JOUR T1 - TibetanGoTinyNet: a light weight U-Net style network for Zero learning of Tibetan Go A1 - Xiali LI A1 - Yanyin ZHANG A1 - Licheng WU A1 - Yandong CHEN A1 - Junzhi YU J0 - Frontiers of Information Technology & Electronic Engineering SP - EP - %@ 2095-9184 Y1 - in press PB - Zhejiang University Press & Springer ER - doi="https://doi.org/10.1631/FITEE.2300493"
Abstract: The game of Tibetan Go faces the scarcity of expert knowledge and studied literature. Therefore, we studied the zero learning model of Tibetan Go under limited computing power resources and proposed a novel scaleinvariant U-Net style two-headed output lightweight network TibetanGoTinyNet. The lightweight convolutional neural networks (CNN) and capsule structure are applied to the encoder and decoder of the network to reduce computational burden and achieve better feature extract results. Several autonomous self-attentive mechanisms are integrated into the network to capture the Tibetan Go boardâĂŹs spatial and global information and select important channels. The training data were generated entirely from self-play games. TibetanGoTinyNet achieved 62%âĂŞ78% winning rates against four other U-Net style models including Ghost-UNet and Res-UNet. It also achieved 75% winning rates in the ablation experiments on the attention mechanism with embedded positional information. The model saved about 33% of the training time with 45%âĂŞ50% winning rates for different Monte Carlo tree search (MCTS) counts when migrated from 9×9 to 11×11 boards. Code for our model is available at https://github.com/paulzyy/TibetanGoTinyNet.
Darkslateblue:Affiliate; Royal Blue:Author; Turquoise:Article
Reference
Open peer comments: Debate/Discuss/Question/Opinion
Open peer comments: Debate/Discuss/Question/Opinion
<1>