# Action-Manipulation Attacks Against Stochastic Bandits: Attacks and Defense

arXiv: https://arxiv.org/abs/2002.08000 · HTML: https://mlsanity.com/paper/2002.08000/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 19 Feb 2020.

## Built on by (most cited first, 8 of 8)

- [Provably Efficient Black-Box Action Poisoning Attacks Against Reinforcement Learning](https://mlsanity.com/paper/2110.04471/) (2021)
- [Understanding the Limits of Poisoning Attacks in Episodic Reinforcement Learning](https://mlsanity.com/paper/2208.13663/) (2022)
- [Efficient Action Poisoning Attacks on Linear Contextual Bandits](https://mlsanity.com/paper/2112.05367/) (2021)
- [Stealthy Adversarial Attacks on Stochastic Multi-Armed Bandits](https://mlsanity.com/paper/2402.13487/) (2024)
- [Optimal Stochastic Nonconvex Optimization with Bandit Feedback](https://mlsanity.com/paper/2103.16082/) (2021)
- [Efficient Preference Poisoning Attack on Offline RLHF](https://mlsanity.com/paper/2605.02495/) (2026)
- [Generalized Likelihood Ratio Test for Adversarially Robust Hypothesis Testing](https://mlsanity.com/paper/2112.02209/) (2021)
- [Reward Teaching for Federated Multi-armed Bandits](https://mlsanity.com/paper/2305.02441/) (2023)

## Abstract (arXiv)

Due to the broad range of applications of stochastic multi-armed bandit model, understanding the effects of adversarial attacks and designing bandit algorithms robust to attacks are essential for the safe applications of this model. In this paper, we introduce a new class of attack named action-manipulation attack. In this attack, an adversary can change the action signal selected by the user. We show that without knowledge of mean rewards of arms, our proposed attack can manipulate Upper Confidence Bound (UCB) algorithm, a widely used bandit algorithm, into pulling a target arm very frequently by spending only logarithmic cost. To defend against this class of attacks, we introduce a novel algorithm that is robust to action-manipulation attacks when an upper bound for the total attack cost is given. We prove that our algorithm has a pseudo-regret upper bounded by O(maxlog T,A), where T is the total number of rounds and A is the upper bound of the total attack cost.