On the Robustness of Epoch-Greedy in Multi-Agent Contextual Bandit Mechanisms

TL;DR

This paper demonstrates that the epsilon-greedy algorithm, a popular method in contextual bandits, is robust against strategic manipulations and adversarial corruptions, maintaining performance despite multiple challenges in multi-agent mechanisms.

Contribution

It extends epsilon-greedy to multi-agent contextual bandit settings and proves its robustness to adversarial corruptions, addressing multiple challenges simultaneously.

Findings

Epsilon-greedy can be adapted for strategic multi-agent environments.

The algorithm's performance degrades linearly with corruption levels.

Epsilon-greedy shows inherent robustness to adversarial attacks.

Abstract

Efficient learning in multi-armed bandit mechanisms such as pay-per-click (PPC) auctions typically involves three challenges: 1) inducing truthful bidding behavior (incentives), 2) using personalization in the users (context), and 3) circumventing manipulations in click patterns (corruptions). Each of these challenges has been studied orthogonally in the literature; incentives have been addressed by a line of work on truthful multi-armed bandit mechanisms, context has been extensively tackled by contextual bandit algorithms, while corruptions have been discussed via a recent line of work on bandits with adversarial corruptions. Since these challenges co-exist, it is important to understand the robustness of each of these approaches in addressing the other challenges, provide algorithms that can handle all simultaneously, and highlight inherent limitations in this combination. In this work, we show that the most prominent contextual bandit algorithm, $\epsilon$-greedy can be extended to handle the challenges introduced by strategic arms in the contextual multi-arm bandit mechanism setting. We further show that $\epsilon$-greedy is inherently robust to adversarial data corruption attacks and achieves performance that degrades linearly with the amount of corruption.