Abstract

We consider binary and multi-class classification problems using hypothesisclasses of neural networks. For a given hypothesis class, we use Rademachercomplexity estimates and direct approximation theorems to obtain a priori errorestimates for regularized loss functionals.

 

Quick Read (beta)