Return to Colloquia & Seminar listing
A geometric perspective on adversarial perturbations
Student-Run Geometry/Topology Seminar| Speaker: | Tommy Hogan, UC Davis |
| Related Webpage: | https://www.math.ucdavis.edu/~tahogan/ |
| Location: | 2112 MSB |
| Start time: | Thu, Oct 18 2018, 12:10PM |
Adversarial examples are slightly perturbed images resembling natural images but maliciously crafted to fool pre-trained image classifiers (usually a deep neural network). We will discuss some geometric explanations on why these examples exist, as well as some algorithms for finding them.