Fredholm integral equations for the training of shallow neural networks

by Patrick Gelß (Zuse Institut Berlin)

Tuesday 27 Jun 2023, 14:30 → 16:30 Europe/Berlin

Seminar room 5 (Villa) (Zeuthen)

We present a novel approach for the training of single-hidden-layer neural networks, based on the approximate solution of associated Fredholm integral equations of the 1. kind by Ritz-Galerkin methods. We show how quantum-inspired tensor formats and Tikhonov regularization can be used to construct continuous counterparts of discrete neural networks with an infinitely large hidden layer. The efficiency and reliability of the introduced approach is illustrated by the practical application to several supervised learning problems.