论文标题
添加剂对抗性攻击和防御的游戏理论分析
A Game Theoretic Analysis of Additive Adversarial Attacks and Defenses
论文作者
论文摘要
对抗性学习的研究是在提出攻击的攻击者和防守者之间的猫和鼠标游戏之后，通过新的防御力来减轻他们的攻击，随后提出了新的攻击，以打破较早的防御力，依此类推。但是，目前尚不清楚是否有没有提出更好的攻击或防御措施的条件。在本文中，我们提出了一个游戏理论框架，用于研究以平衡状态存在的攻击和防御措施。在基础二进制分类器的局部线性决策边界模型下，我们证明了快速梯度方法攻击和随机平滑防御构成NASH平衡。然后，我们显示如何从数据生成分布中有限的许多样本来估算这种平衡防御，并为我们的近似性能得出概括。
Research in adversarial learning follows a cat and mouse game between attackers and defenders where attacks are proposed, they are mitigated by new defenses, and subsequently new attacks are proposed that break earlier defenses, and so on. However, it has remained unclear as to whether there are conditions under which no better attacks or defenses can be proposed. In this paper, we propose a game-theoretic framework for studying attacks and defenses which exist in equilibrium. Under a locally linear decision boundary model for the underlying binary classifier, we prove that the Fast Gradient Method attack and the Randomized Smoothing defense form a Nash Equilibrium. We then show how this equilibrium defense can be approximated given finitely many samples from a data-generating distribution, and derive a generalization bound for the performance of our approximation.
