Seminario - On two level and random preconditioning in machine learning

il 7 settembre p.v. alle ore 11:00 nell'aula 36 del  DMI, nell'ambito delle attività del CIMAT (Centro Interdipartimentale diMatematica per la Tecnologia "A. M. Anile"), il prof. Alfio Borzì dell'Università di Würzburg terrà un seminario dal titolo "On two level and random preconditioning in machine learning"

Abstract

Neural networks are machine learning models for constructing universal function approximation algorithms. The learning of these networks requires to iteratively solve large nonlinear optimization problems, and gradient methods play a central role in this framework, which motivates the investigation of techniques for accelerating and robustifying these methods.

In this talk, a twolevel- and gradient-based learning scheme and a regularized preconditioned gradient method are discussed and applied to the training of multi-layer neural networks. In the former method, the tuning of the gradient and the coarse-level correction scheme are based on the knowledge of the Hessian of the loss function. In the latter case, a suitable regularization and a preconditioner based on random normal projections are analysed.