Stealthy Adversarial Attacks on Stochastic Multi-Armed Bandits
DOI:
https://doi.org/10.1609/aaai.v38i14.29506Keywords:
ML: Online Learning & Bandits, ML: Adversarial Learning & RobustnessAbstract
Adversarial attacks against stochastic multi-armed bandit (MAB) algorithms have been extensively studied in the literature. In this work, we focus on reward poisoning attacks and find most existing attacks can be easily detected by our proposed detection method based on the test of homogeneity, due to their aggressive nature in reward manipulations. This motivates us to study the notion of stealthy attack against stochastic MABs and investigate the resulting attackability. Our analysis shows that against two popularly employed MAB algorithms, UCB1 and $\epsilon$-greedy, the success of a stealthy attack depends on the environmental conditions and the realized reward of the arm pulled in the first round. We also analyze the situation for general MAB algorithms equipped with our attack detection method and find that it is possible to have a stealthy attack that almost always succeeds. This brings new insights into the security risks of MAB algorithms.Downloads
Published
2024-03-24
How to Cite
Wang, Z., Wang, H., & Wang, H. (2024). Stealthy Adversarial Attacks on Stochastic Multi-Armed Bandits. Proceedings of the AAAI Conference on Artificial Intelligence, 38(14), 15770-15777. https://doi.org/10.1609/aaai.v38i14.29506
Issue
Section
AAAI Technical Track on Machine Learning V