Cross-Domain Transferability of Adversarial Perturbations - CAP6412 Spring 2021

Cross-Domain Transferability of Adversarial Perturbations - CAP6412 Spring 2021

Explore the concept of cross-domain transferability of adversarial perturbations in this 42-minute lecture from the University of Central Florida. Delve into the paper's contents, covering the introduction, related work, and the Transferable Generative Adversarial Model. Examine key components such as Discriminator Loss and Relativistic Cross-Entropy. Analyze experimental settings and results, including training progress visualization, Gaussian kernel size, and attention shift. Conclude with a discussion of arguments for and against the presented approach, gaining a comprehensive understanding of this advanced topic in adversarial machine learning.