On The Topological Expressive Power of Neural Networks

Giovanni Petri, António Leitão

Research output: Contribution to conferencePosterpeer-review

17 Downloads (Pure)

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 architecture as a measure of its intrinsic expressive power.
By sampling a large number of neural architectures 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.
Original languageEnglish
Number of pages1
Publication statusPublished - 11 Dec 2020
EventTopological Data Analysis and Beyond. : Workshop at NeurIPS 2020 - Virtual
Duration: 11 Dec 202011 Dec 2020
Conference number: 2020
https://tda-in-ml.github.io/

Conference

ConferenceTopological Data Analysis and Beyond.
Abbreviated titleNeurIPS
Period11/12/2011/12/20
Internet address

Keywords

  • Neural Networks
  • Expressive Power
  • Decision Boundary
  • Classification

Fingerprint

Dive into the research topics of 'On The Topological Expressive Power of Neural Networks'. Together they form a unique fingerprint.

Cite this