The University of Arizona
Please note that this event has ended!

Adversarial Neighborhoods: Statistical analysis of adversarial attacks against image classification models

Program in Applied Mathematics Brown Bag Seminar

Adversarial Neighborhoods: Statistical analysis of adversarial attacks against image classification models
Series: Program in Applied Mathematics Brown Bag Seminar
Location: ONLINE
Presenter: Brian Bell, Program in Applied Mathematics, University of Arizona

Artificial Neural Networks have achieved renowned success in automated image classification and many other tasks. These models are vulnerable to small "Adversarial" perturbations on samples which greatly disrupt classification output. In this work we describe methods used to generate adversarial examples and characteristics of these examples. We will also introduce statistical analysis of these examples and propose a definition that can be used to characterize and identify such examples.

Zoom: https://arizona.zoom.us/j/98593835992     Password: 021573