Can Deep Reinforcement Learning Solve Erdos-Selfridge-Spencer Games?

Maithra Raghu, Alex Irpan, Jacob Andreas, Bobby Kleinberg, Quoc Le, Jon Kleinberg
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:4238-4246, 2018.

Abstract

Deep reinforcement learning has achieved many recent successes, but our understanding of its strengths and limitations is hampered by the lack of rich environments in which we can fully characterize optimal behavior, and correspondingly diagnose individual actions against such a characterization. Here we consider a family of combinatorial games, arising from work of Erdos, Selfridge, and Spencer, and we propose their use as environments for evaluating and comparing different approaches to reinforcement learning. These games have a number of appealing features: they are challenging for current learning approaches, but they form (i) a low-dimensional, simply parametrized environment where (ii) there is a linear closed form solution for optimal behavior from any state, and (iii) the difficulty of the game can be tuned by changing environment parameters in an interpretable way. We use these Erdos-Selfridge-Spencer games not only to compare different algorithms, but test for generalization, make comparisons to supervised learning, analyse multiagent play, and even develop a self play algorithm.

Cite this Paper

BibTeX
@InProceedings{pmlr-v80-raghu18a, title = {Can Deep Reinforcement Learning Solve {E}rdos-{S}elfridge-{S}pencer Games?}, author = {Raghu, Maithra and Irpan, Alex and Andreas, Jacob and Kleinberg, Bobby and Le, Quoc and Kleinberg, Jon}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {4238--4246}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/raghu18a/raghu18a.pdf}, url = {https://proceedings.mlr.press/v80/raghu18a.html}, abstract = {Deep reinforcement learning has achieved many recent successes, but our understanding of its strengths and limitations is hampered by the lack of rich environments in which we can fully characterize optimal behavior, and correspondingly diagnose individual actions against such a characterization. Here we consider a family of combinatorial games, arising from work of Erdos, Selfridge, and Spencer, and we propose their use as environments for evaluating and comparing different approaches to reinforcement learning. These games have a number of appealing features: they are challenging for current learning approaches, but they form (i) a low-dimensional, simply parametrized environment where (ii) there is a linear closed form solution for optimal behavior from any state, and (iii) the difficulty of the game can be tuned by changing environment parameters in an interpretable way. We use these Erdos-Selfridge-Spencer games not only to compare different algorithms, but test for generalization, make comparisons to supervised learning, analyse multiagent play, and even develop a self play algorithm.} }
Endnote
%0 Conference Paper %T Can Deep Reinforcement Learning Solve Erdos-Selfridge-Spencer Games? %A Maithra Raghu %A Alex Irpan %A Jacob Andreas %A Bobby Kleinberg %A Quoc Le %A Jon Kleinberg %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-raghu18a %I PMLR %P 4238--4246 %U https://proceedings.mlr.press/v80/raghu18a.html %V 80 %X Deep reinforcement learning has achieved many recent successes, but our understanding of its strengths and limitations is hampered by the lack of rich environments in which we can fully characterize optimal behavior, and correspondingly diagnose individual actions against such a characterization. Here we consider a family of combinatorial games, arising from work of Erdos, Selfridge, and Spencer, and we propose their use as environments for evaluating and comparing different approaches to reinforcement learning. These games have a number of appealing features: they are challenging for current learning approaches, but they form (i) a low-dimensional, simply parametrized environment where (ii) there is a linear closed form solution for optimal behavior from any state, and (iii) the difficulty of the game can be tuned by changing environment parameters in an interpretable way. We use these Erdos-Selfridge-Spencer games not only to compare different algorithms, but test for generalization, make comparisons to supervised learning, analyse multiagent play, and even develop a self play algorithm.
APA
Raghu, M., Irpan, A., Andreas, J., Kleinberg, B., Le, Q. & Kleinberg, J.. (2018). Can Deep Reinforcement Learning Solve Erdos-Selfridge-Spencer Games?. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:4238-4246 Available from https://proceedings.mlr.press/v80/raghu18a.html.

Related Material