Bob Williamson: Beyond Expectations Better than Average: Foundations of ML Systems

Bob Williamson is Professor for Foundations of Machine Learning Systems at the University of Tübingen. This talk was part of the colloquium of the Cluster of Excellence Machine Learning: New Perspectives for Science. Abstract: Machine Learning (ML) technology provides value by its integration into a system, either technological or sociotechnical. This requires an external outwardlooking view of ML, rather than the usual internal inwardlooking view of much ML research. I will illustrate the external view by considering some foundational concerns regarding ML which motivates the research I propose to conduct in Tübingen. The starting point turns out to be a more careful analysis of expectation or averages. Somewhat astonishingly, there is a lot to be said, and a lot to be yet discovered, about this apparently elementary building block of ML systems, and about its natural generalisations. I will argue why one should consider moving beyond expectations, and explore some of the intriguing technical probl