Adversarial Policy Learning in Two-player Competitive Games
Wenbo Guo, Xian Wu, Sui Huang, Xinyu Xing
Abstract
In a two-player deep reinforcement learning task, recent work shows an attacker could learn an adversarial policy that triggers a target agent to perform poorly and even react in an undesired way. However, its efficacy heavily relies upon the zero-sum assumption made in the two-player game. In this work, we propose a new adversarial learning algorithm. It addresses the problem by resetting the optimization goal in the learning process and designing a new surrogate optimization function. Our experiments show that our method significantly improves adversarial agents’ exploitability compared with the state-of-art attack. Besides, we also discover that our method could augment an agent with the ability to abuse the target game’s unfairness. Finally, we show that agents adversarially re-trained against our adversarial agents could obtain stronger adversary-resistance.
BibTeX
@InProceedings{pmlr-v139-guo21b,
title = {Adversarial Policy Learning in Two-player Competitive Games},
author = {Guo, Wenbo and Wu, Xian and Huang, Sui and Xing, Xinyu},
booktitle = {Proceedings of the 38th International Conference on Machine Learning},
pages = {3910--3919},
year = {2021},
editor = {Meila, Marina and Zhang, Tong},
volume = {139},
series = {Proceedings of Machine Learning Research},
month = {18--24 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v139/guo21b/guo21b.pdf},
url = {https://proceedings.mlr.press/v139/guo21b.html},
abstract = {In a two-player deep reinforcement learning task, recent work shows an attacker could learn an adversarial policy that triggers a target agent to perform poorly and even react in an undesired way. However, its efficacy heavily relies upon the zero-sum assumption made in the two-player game. In this work, we propose a new adversarial learning algorithm. It addresses the problem by resetting the optimization goal in the learning process and designing a new surrogate optimization function. Our experiments show that our method significantly improves adversarial agents’ exploitability compared with the state-of-art attack. Besides, we also discover that our method could augment an agent with the ability to abuse the target game’s unfairness. Finally, we show that agents adversarially re-trained against our adversarial agents could obtain stronger adversary-resistance.}
}