Some thoughts and questions towards a statistical understanding of DNNs

So far, our statistical understanding of the learning mechanisms of deep neural networks. (DNNs) is rather limited. Part of the reasons for this lack of understanding is the fact
that in many cases the tools of classical statistical learning theory can no longer be applied. In this talk, I will present some thoughts and possible questions that may be relevant for a successful end-to-end analysis of DNNs. In particular, I will discuss the role of initialization, overparameterization, global minima, and optimization procedures.

***Due to the current pandemic this colloquium will be conducted online. We invite you to join and spread the news. In case you wish to participate, please send an email up front to liv.heinecke[at]uni-potsdam.de to receive the zoom invitation link***