Spencer Frei

Deep learning's success has revealed a number of phenomena that appear to conflict with classical inuitions in the fields of optimization and statistics. First, the objective functions formulated in deep learning are highly nonconvex but are...