Stealthy Adversarial Attacks on Stochastic Multi-Armed Bandits

arXiv:2402.13487v1 Announce Type: cross
Abstract: 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 …

adversarial adversarial attacks algorithms arxiv attacks bandit can cs.cr cs.lg detection find focus literature poisoning poisoning attacks reward test work