%0 Report %1 yamamoto2018learning %A Yamamoto, Masahito %A Kato, Shu %A Iizuka, Hiroyuki %D 2018 %K Curling %T Learning of Expected Scores Distribution for Positions of Digital Curling %U https://ipsj.ixsq.nii.ac.jp/ej/?action=repository_action_common_download&item_id=190730&item_no=1&attribute_id=1&file_no=1 %X Curling is a team sport played by two teams of four players on the ice. Recently, not only the skill but also the strategy becomes very important for winning the curling games. In order to discuss about the optimal strategy for curling, the curling playing simulator called by digital curling has been developed by Ito and his colleagues. In digital curling, rules are equivalent to actual curling, and the outcome of the delivery is simulated by internal physical computation. On the digital curling, we have developed curling AI program jiritsu-kun which can search for the best play based on the game tree search. In this paper, we propose the learning method of the expected scores distribution at the end of the ”end” as a static evaluation function of the game tree search. It is based on a deep neural network model. In order to evaluate our proposed method, we compare the learned evaluation function with hand-crafted evaluation function in the previous version of jiritsu-kun.

@techreport{yamamoto2018learning, abstract = {Curling is a team sport played by two teams of four players on the ice. Recently, not only the skill but also the strategy becomes very important for winning the curling games. In order to discuss about the optimal strategy for curling, the curling playing simulator called by digital curling has been developed by Ito and his colleagues. In digital curling, rules are equivalent to actual curling, and the outcome of the delivery is simulated by internal physical computation. On the digital curling, we have developed curling AI program jiritsu-kun which can search for the best play based on the game tree search. In this paper, we propose the learning method of the expected scores distribution at the end of the ”end” as a static evaluation function of the game tree search. It is based on a deep neural network model. In order to evaluate our proposed method, we compare the learned evaluation function with hand-crafted evaluation function in the previous version of jiritsu-kun.}, added-at = {2024-04-29T18:50:48.000+0200}, author = {Yamamoto, Masahito and Kato, Shu and Iizuka, Hiroyuki}, biburl = {https://www.bibsonomy.org/bibtex/215f2d33690e66f471396f0d9e3154fe8/cckonstanz}, institution = {Information Processing Society of Japan}, interhash = {328b1feba48745fec7c2a95399045443}, intrahash = {15f2d33690e66f471396f0d9e3154fe8}, keywords = {Curling}, timestamp = {2024-04-29T18:50:48.000+0200}, title = {Learning of Expected Scores Distribution for Positions of Digital Curling}, url = {https://ipsj.ixsq.nii.ac.jp/ej/?action=repository_action_common_download&item_id=190730&item_no=1&attribute_id=1&file_no=1}, year = 2018 }