From the EIC: Robust Machine Learning

Jörg Henkel

Abstract: fig orientation="portrait" position="float" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <graphic orientation="portrait" position="float" xlink:href="henke-2984228.tif"/> </fig> <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Machine learning techniques</b> have become pervasive through many technical fields but an obstacle for employment is often the criterion of robustness. While machine learning can be a great means to improve upon the quality of traditional optimization techniques in uncritical scenarios (e.g., a customized online search result that proposes to a consumer a more or less well-fitting new product advertisement), it may be prohibitive to employ when directly embedded in critical decision flows (e.g., a self-driving car that needs to decide whether to engage an emergency brake). In the latter case, robustness is one mandatory constraint. Robustness can have many facets; some of them are covered by this timely special issue that represents the state of the art from a design and test point of view. Many thanks to the Guest Editors Theocharis Theocharides, Muhammad Shafique, Jungwook Choi, and Onur Mutlu for editing this special issue that includes two keynote articles, a survey on “Robust Machine Learning Systems: Challenges, Current Trends, Perspectives, and the Road Ahead,” and six technical articles.