Learning Strategy-Aware Linear Classifiers

TL;DR

This paper investigates the challenges of learning linear classifiers in strategic settings where agents attempt to game the system, revealing fundamental incompatibilities between regret measures and proposing a strategy-aware algorithm with theoretical guarantees.

Contribution

It introduces a strategy-aware algorithm for minimizing Stackelberg regret in strategic classification, with proven near-optimal bounds and insights into regret measure incompatibilities.

Findings

Stackelberg and external regret are strongly incompatible in strategic classification.

The proposed algorithm achieves nearly matching upper and lower regret bounds.

Simulations support the theoretical analysis and demonstrate practical effectiveness.

Abstract

We address the question of repeatedly learning linear classifiers against agents who are strategically trying to game the deployed classifiers, and we use the Stackelberg regret to measure the performance of our algorithms. First, we show that Stackelberg and external regret for the problem of strategic classification are strongly incompatible: i.e., there exist worst-case scenarios, where any sequence of actions providing sublinear external regret might result in linear Stackelberg regret and vice versa. Second, we present a strategy-aware algorithm for minimizing the Stackelberg regret for which we prove nearly matching upper and lower regret bounds. Finally, we provide simulations to complement our theoretical analysis. Our results advance the growing literature of learning from revealed preferences, which has so far focused on "smoother" assumptions from the perspective of the learner and the agents respectively.