Adversarial Attacks on Stochastic Bandits
By Kwang-Sung Jun, Lihong Li, Xiaojin Zhu. 2018
139 citations recorded by the source corpus.
We study adversarial attacks that manipulate the reward signals to control the actions chosen by a stochastic multi-armed bandit algorithm. We propose the first attack against two popular bandit algorithms: $\epsilon$-greedy and UCB, \emph{without} knowledge of the mean rewards. The attacker is able to spend only logarithmic effort, multiplied by a problem-specific parameter that becomes smaller as the bandit problem gets easier to attack. The result means the attacker can easily hijack the behavior of the bandit algorithm to promote or obstruct certain actions, say, a particular medical treatment. As bandits are seeing increasingly wide use in practice, our study exposes a significant security threat.
Subfields: Artificial Intelligence, Management Science and Operations Research
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Xuezhou Zhang
- Justin Hsu
- Xiaojin Zhu
- Shike Mei
- Daniel Kifer
- Danfeng Zhang
- Oisin Mac Aodha
- Percy Liang
In Management Science and Operations Research
Citations in the corpus
- Data Poisoning against Differentially-Private Learners: Attacks and Defenses (2019, background): “e is adversarial examples [20], where inputs are crafted to fool classifiers. More subtle methods aim to recover training examples [9], or even extract the model itself [21]. In data poisoning attacks [3, 12, 15, 1, 14, 22, 16, 10, 24, 23], the adversary corrupts examples at training time to manipula”
- Online Data Poisoning Attack (2019, background): “A variety of attacks against other learners have been developed, including neural networks [20, 12], autoregressive models [7, 1], linear and stochastic bandits [17, 11], collaborative filtering [14], and sentiment analysis [21], etc.”