When Should You Defend Your Classifier -- A Game-theoretical Analysis of Countermeasures against Adversarial Examples
This work addresses the need for more realistic evaluation scenarios in adversarial machine learning, though it is incremental by refining existing game-theoretical approaches.
The paper tackles the problem of unrealistic assumptions in evaluating adversarial machine learning defenses by proposing a game-theoretical model that incorporates economic costs and accuracy trade-offs, finding that the maximum amount of adversarial examples is the most influential factor in practical settings.
Adversarial machine learning, i.e., increasing the robustness of machine learning algorithms against so-called adversarial examples, is now an established field. Yet, newly proposed methods are evaluated and compared under unrealistic scenarios where costs for adversary and defender are not considered and either all samples or no samples are adversarially perturbed. We scrutinize these assumptions and propose the advanced adversarial classification game, which incorporates all relevant parameters of an adversary and a defender. Especially, we take into account economic factors on both sides and the fact that all so far proposed countermeasures against adversarial examples reduce accuracy on benign samples. Analyzing the scenario in detail, where both players have two pure strategies, we identify all best responses and conclude that in practical settings, the most influential factor might be the maximum amount of adversarial examples.