A lot of academic attention has been focused on the topic of adversarial attacks against neural networks recently, and for good reason. Because we still lack a thorough understanding of why neural networks work well, the failure modes of neural networks are quite surprising1.
