Brendon and Anish interview Zach Trefler, a student at the University of Toronto, about his research on designing neural networks to create adversarial examples to defeat, and then improve the security of, a voiceprinting system.

We discuss how generative-adversarial networks are applied and how they can be applied to generate adversarial examples, in order to trick a security system into classifying the examples as a completely different user, and then how this can be used to develop a more secure system.

Please send comments to .