Constructing Unrestricted Adversarial Examples with Generative Models

• Published in: arXiv.org
• Main Authors: Yang, Song, Shu, Rui, Kushman, Nate, Ermon, Stefano
• Format: Paper
• Language: English
• Published: Ithaca Cornell University Library, arXiv.org 02.12.2018
• Subjects: Classifiers; Conditioning; Data points

Adversarial examples are typically constructed by perturbing an existing data point within a small matrix norm, and current defense methods are focused on guarding against this type of attack. In this paper, we propose unrestricted adversarial examples, a new threat model where the attackers are not...