Adversarial Uncertainty Learning in Deep Neural Networks

Uncertainty estimation in deep neural network models has become a mainstream research topic in recent years. However, little to no effort has been made to assess the robustness of the estimated uncertainty. One such assessment can be based on white-box adversarial attacks. In my talk I will give a short introduction to uncertainty estimation for deep networks and to adversarial learning. Then, I will present my results on attacking a system capable of uncertainty estimation during its decision making process. Finally, I will show initial results on adversarial model training in order to minimize the vulnerability of a model to such attacks.

---------------------------------------

Plan wystąpień w tej edycji jest dostępny tutaj

The schedule of presentations can be checked here