Some Results on the Expressivity of Deep Neural Networks

Document Type

Dissertation

Abstract

Deep Neural Networks (NNs), have attracted a lot of attention since the year 2000, thanks to their impressive state-of-the-art performance in the machine learning field and its tasks. In this thesis, we study the expressivity of deep neural networks, especially the number of linear regions for PPNNs and Piecewise Polynomial Convolutional Neural Networks (PPCNNs). Our results suggest that, deep PPCNNs have way more of an expressivity than shallower PPCNNs having equal number of parameters; and deep PPCNNs have more expressivity than deep PPNNs under some wild architectural assumptions.

First Page

i

Last Page

27

Publication Date

12-30-2022

Comments

Thesis submitted to the Deanship of Graduate and Postdoctoral Studies

In partial fulfillment of the requirements for the M.Sc degree in Machine Learning

Advisors: Dr. Huan Xiong, Dr. Rao Anwer

Online access provided for MBZUAI patrons

Share

COinS