Guest editors

Leon Bungert - University of Wuerzburg

Des Higham - The University of Edinburgh

Laura Thesing - The Ludwig Maximilian University of Munich 

About

Artificial Intelligence (AI) is now prevalent in high-risk domains such as medical image diagnosis, drug discovery, and self-driving cars. For this reason, the vulnerability of AI systems to adversaries remains a critical concern. Adversarial attacks may take a wide variety of forms, but in essence they reveal inherent instabilities in either the problem specifications or the accompanying algorithms. Hence, they are attractive to adversarial users and they highlight that a system might not always perform as expected. A classical example of an adversarial attack is a perturbation to an image that is imperceptible to the human eye but causes a change in the classification from an AI system. While recent years have seen substantial progress on the algorithmic side of robustifying machine learning, the attack-versus-defence arms race shows no sign of ending. Furthermore, obtaining meaningful mathematical guarantees for the robustness of the resulting models is still a challenging and—to a large extent—open problem. Hurdles that remain include unifying the many different notions of robustness, obtaining reasonable estimates on the sample complexity of the learning and defence methods, and tackling and understanding the possible trade-off between robustness and accuracy. The theme of this special issue—addressing the mathematics of adversarial and robust machine learning—attracts researchers from different branches of applied mathematics, statistics and theoretical computer science, including those with backgrounds in optimization and numerical analysis, high dimensional and stochastic analysis, complexity theory, formal reasoning and software testing. By bringing together a range of viewpoints in a unified manner, this special issue will be both timely and impactful. 

Deadline

31 December 2024 

Instructions for authors

Please visit our instructions for authors page detailing how to submit your paper.