Holger Hoos (RWTH Aachen University)

Title: News from the frontier of neural network robustness verification
Abstract: Over the last decade, machine learning methods, notably neural networks, have played a key role in enabling major progress in artificial intelligence and its applications. Unfortunately, neural networks are known to be sensitive to input perturbations, including adversarial attacks. In this talk, I will give an introduction to neural network robustness and an overview of the work on formal characterisation and verification of neural network robustness in my group at RWTH Aachen and Leiden University. I will explain how work in this area nicely and effectively combines a broad range of methods across the field of AI, including machine learning, optimisation and automated reasoning.