In recent years, the remarkable performance of deep neural networks has led to their application across various safety-critical domains. However, their vulnerability to adversarial examples—small, often imperceptible input perturbations that cause incorrect outputs—has raised significant concerns about their reliability.
Therefore, an important subfield of machine learning that aims to assess and improve a model's robustness has emerged. Researchers are developing empirical and formal methods to assess a model's resilience and design inherently robust architectures capable of withstanding adversarial attacks.
In this seminar, we will explore key aspects of robust machine learning, including adversarial attack strategies, techniques for formal verification, and robust training methods.
The seminar is held in English.
Registration to the seminar is handled via the SuPra system.
Prof. Dr. Holger H. Hoos
Chair Holder, Alexander von Humboldt Professor
E-mail: hh[at]aim[dot]rwth-aachen[dot]de
Phone: +49 241 80 21451
