ELF OpenGo: an analysis and open reimplementation of AlphaZero

Yuandong Tian, Jerry Ma, Qucheng Gong, Shubho Sengupta, Zhuoyuan Chen, James Pinkerton, Larry Zitnick
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:6244-6253, 2019.

Abstract

The AlphaGo, AlphaGo Zero, and AlphaZero series of algorithms are remarkable demonstrations of deep reinforcement learning’s capabilities, achieving superhuman performance in the complex game of Go with progressively increasing autonomy. However, many obstacles remain in the understanding of and usability of these promising approaches by the research community. Toward elucidating unresolved mysteries and facilitating future research, we propose ELF OpenGo, an open-source reimplementation of the AlphaZero algorithm. ELF OpenGo is the first open-source Go AI to convincingly demonstrate superhuman performance with a perfect (20:0) record against global top professionals. We apply ELF OpenGo to conduct extensive ablation studies, and to identify and analyze numerous interesting phenomena in both the model training and in the gameplay inference procedures. Our code, models, selfplay datasets, and auxiliary data are publicly available.

Cite this Paper

BibTeX
@InProceedings{pmlr-v97-tian19a, title = {{ELF} {O}pen{G}o: an analysis and open reimplementation of {A}lpha{Z}ero}, author = {Tian, Yuandong and Ma, Jerry and Gong, Qucheng and Sengupta, Shubho and Chen, Zhuoyuan and Pinkerton, James and Zitnick, Larry}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {6244--6253}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, month = {09--15 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v97/tian19a/tian19a.pdf}, url = {https://proceedings.mlr.press/v97/tian19a.html}, abstract = {The AlphaGo, AlphaGo Zero, and AlphaZero series of algorithms are remarkable demonstrations of deep reinforcement learning’s capabilities, achieving superhuman performance in the complex game of Go with progressively increasing autonomy. However, many obstacles remain in the understanding of and usability of these promising approaches by the research community. Toward elucidating unresolved mysteries and facilitating future research, we propose ELF OpenGo, an open-source reimplementation of the AlphaZero algorithm. ELF OpenGo is the first open-source Go AI to convincingly demonstrate superhuman performance with a perfect (20:0) record against global top professionals. We apply ELF OpenGo to conduct extensive ablation studies, and to identify and analyze numerous interesting phenomena in both the model training and in the gameplay inference procedures. Our code, models, selfplay datasets, and auxiliary data are publicly available.} }
Endnote
%0 Conference Paper %T ELF OpenGo: an analysis and open reimplementation of AlphaZero %A Yuandong Tian %A Jerry Ma %A Qucheng Gong %A Shubho Sengupta %A Zhuoyuan Chen %A James Pinkerton %A Larry Zitnick %B Proceedings of the 36th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Ruslan Salakhutdinov %F pmlr-v97-tian19a %I PMLR %P 6244--6253 %U https://proceedings.mlr.press/v97/tian19a.html %V 97 %X The AlphaGo, AlphaGo Zero, and AlphaZero series of algorithms are remarkable demonstrations of deep reinforcement learning’s capabilities, achieving superhuman performance in the complex game of Go with progressively increasing autonomy. However, many obstacles remain in the understanding of and usability of these promising approaches by the research community. Toward elucidating unresolved mysteries and facilitating future research, we propose ELF OpenGo, an open-source reimplementation of the AlphaZero algorithm. ELF OpenGo is the first open-source Go AI to convincingly demonstrate superhuman performance with a perfect (20:0) record against global top professionals. We apply ELF OpenGo to conduct extensive ablation studies, and to identify and analyze numerous interesting phenomena in both the model training and in the gameplay inference procedures. Our code, models, selfplay datasets, and auxiliary data are publicly available.
APA
Tian, Y., Ma, J., Gong, Q., Sengupta, S., Chen, Z., Pinkerton, J. & Zitnick, L.. (2019). ELF OpenGo: an analysis and open reimplementation of AlphaZero. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:6244-6253 Available from https://proceedings.mlr.press/v97/tian19a.html.

Related Material