TY - THES A1 - Habermayr, Daniel T1 - Visualizing GAN-based Extreme Learned Image Compression N2 - Systems based on Artificial Intelligence are thrusting their way into evermore fields of application, with the potential to globally revolutionize working and living environments over coming decades. Especially in highly sensitive use-cases with lots of responsibility, like medicine, law enforcement or military, it is of utmost importance to be able to fully comprehend why and how an artificially intelligent program acts. The field of Explainable AI addresses just this task, developing techniques to gain insight into otherwise hard to interpret machine learning models. Knowledge gained through these explanations can also be used to debug and improve any existing models from all kinds of application fields. In this thesis, several appropriate Explainable AI techniques are used to investigate a Deep Compression System for images based on Generative Adversarial Networks. KW - Generative Adversarial Network KW - Deep Compression KW - Explainable AI KW - Künstliche Intelligenz Y1 - 2021 U6 - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:860-opus4-2990 ER -