Seminar

Abstract (nur in Englisch verfügbar): 

In recent years, computer vision models have achieved increased performance on various recognition tasks. These models are vulnerable against adversarial examples. A small perturbation in the input image is enough for misclassification. Adversarial examples reveals that models perform differently with the human visual system, and provides opportunities for understanding and improving these models. Therefore, in this talk an overview of SOTA attacks and defenses will be given.