Robust Thompson Sampling Algorithms Against Reward Poisoning Attacks

TL;DR

This paper develops robust Thompson sampling algorithms that maintain near-optimal performance in online decision-making tasks even when facing adversarial reward poisoning attacks, by using pseudo-posteriors to mitigate manipulation.

Contribution

The authors introduce novel robust Thompson sampling algorithms for stochastic and contextual linear bandits that are effective against reward poisoning, regardless of attacker awareness.

Findings

Guarantee near-optimal regret under any attack strategy.

Propose pseudo-posteriors to reduce manipulation impact.

Applicable in both attacker-aware and unaware scenarios.

Abstract

Thompson sampling is one of the most popular learning algorithms for online sequential decision-making problems and has rich real-world applications. However, current Thompson sampling algorithms are limited by the assumption that the rewards received are uncorrupted, which may not be true in real-world applications where adversarial reward poisoning exists. To make Thompson sampling more reliable, we want to make it robust against adversarial reward poisoning. The main challenge is that one can no longer compute the actual posteriors for the true reward, as the agent can only observe the rewards after corruption. In this work, we solve this problem by computing pseudo-posteriors that are less likely to be manipulated by the attack. We propose robust algorithms based on Thompson sampling for the popular stochastic and contextual linear bandit settings in both cases where the agent is aware or unaware of the budget of the attacker. We theoretically show that our algorithms guarantee near-optimal regret under any attack strategy.