Topological Expressive Power of Neural Networks

Abstract

We propose a topological description of neural network expressive power. We adopt the topology of the space of decision boundaries realized by a neural network as a measure of its intrinsic expressive power. By sampling a large number of neural arhitectures with different sizes and design, we show how such measure of expressive power depends on the properties of the architectures like depth, width and other related quantities.

Publication
In NeurIPS 2020 Workshop on Topological Data Analysis and Beyond
Avatar
António Leitão
Student

Topological analysis of learning systems.