"Some thoughts and questions towards a statistical understanding of DNNs" is the title of Prof. Steinwart's lecture.
So far, our statistical understanding of deep neural networks (DNNs) is rather limited, in particular if over-parametrized DNNs are considered. Part of the reasons for this lack of understanding is the fact that forsuch large DNNs the tools of classical statistical learning theory can no longer be applied. Nonetheless,over-parametrized DNNs are typically performing well in practice. In this talk, I will discuss some recentresult and resulting possible questions that may be relevant for a successful end-to-end analysis of DNNs.
Joint Work with Nicole Mücke and David Holzmüller