betahaus - innospace, Prinzessinnenstrasse 19-20 10969, Berlin
Talk 1: Katharina Rasch - Fooling neural networks
Abstract: It is surprisingly easy to fool deep neural networks (and other machine learning models). Change a few pixels in an image (changes not visible to humans!) and suddenly a zebra is classified with high confidence as a microwave. Let's look at some methods for generating such adversarial examples and discuss what this could tell us about what our models are actually learning.
Here is a good read if you can't make it.