Workshop „Adversarial Examples“ mit Dr. Fabian Offert (University of California, Santa Barbara)

Beschreibung

Adversarial examples are images that elude their classification by deep neural networks. They exploit certain known weaknesses of these networks — for instance a preference for texture over shape — by introducing invisible changes to an image on the pixel level. Such images can then become instruments for stress-testing artificial intelligence systems, tools of resistance against social surveillance and control, or even media of artistic expression. This workshop introduces participants to the theory and practice of adversarial examples. With only a few lines of code, we will create images that confuse many well-known neural network architectures, and explore the limits of current-generation artificial intelligence systems.