A geometric perspective on information plane analysis

Mina Basirat, Bernhard C. Geiger, Peter M. Roth*

*Korrespondierende/r Autor/-in für diese Arbeit

Publikation: Beitrag in einer FachzeitschriftArtikelBegutachtung

Abstract

Information plane analysis, describing the mutual information between the input and a hidden layer and between a hidden layer and the target over time, has recently been proposed to analyze the training of neural networks. Since the activations of a hidden layer are typically continuous-valued, this mutual information cannot be computed analytically and must thus be estimated, resulting in apparently inconsistent or even contradicting results in the literature. The goal of this paper is to demonstrate how information plane analysis can still be a valuable tool for analyzing neural network training. To this end, we complement the prevailing binning estimator for mutual information with a geometric interpretation. With this geometric interpretation in mind, we evaluate the impact of regularization and interpret phenomena such as underfitting and overfitting. In addition, we investigate neural network learning in the presence of noisy data and noisy labels.

Originalspracheenglisch
Aufsatznummer711
FachzeitschriftEntropy
Jahrgang23
Ausgabenummer6
DOIs
PublikationsstatusVeröffentlicht - Juni 2021

ASJC Scopus subject areas

  • Information systems
  • Mathematische Physik
  • Physik und Astronomie (sonstige)
  • Elektrotechnik und Elektronik

Fingerprint

Untersuchen Sie die Forschungsthemen von „A geometric perspective on information plane analysis“. Zusammen bilden sie einen einzigartigen Fingerprint.

Dieses zitieren