Numerosity Representation in InfoGAN: An Empirical Study

Andrea Zanetti, Alberto Testolin, Marco Zorzi, Pawel Wawrzynski

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

It has been shown that “visual numerosity emerges as a statistical property of images in ‘deep networks’ that learn a hierarchical generative model of the sensory input”, through unsupervised deep learning [1]. The original deep generative model was based on stochastic neurons and, more importantly, on input (image) reconstruction. Statistical analysis highlighted a correlation between the numerosity present in the input and the population activity of some neurons in the second hidden layer of the network, whereas population activity of neurons in the first hidden layer correlated with total area (i.e., number of pixels) of the objects in the image. Here we further investigate whether numerosity information can be isolated as a disentangled factor of variation of the visual input. We train in unsupervised and semi-supervised fashion a latent-space generative model that has been shown capable of disentangling relevant semantic features in a variety of complex datasets, and we test its generative performance under different conditions. We then propose an approach to the problem based on the assumption that, in order to let numerosity emerge as disentangled factor of variation, we need to cancel out the sources of variation at graphical level.

Original languageEnglish
Title of host publicationAdvances in Computational Intelligence - 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Proceedings
EditorsIgnacio Rojas, Gonzalo Joya, Andreu Catala
PublisherSpringer Verlag
Pages49-60
Number of pages12
ISBN (Print)9783030205171
DOIs
Publication statusPublished - Jan 1 2019
Event15th International Work-Conference on Artificial Neural Networks, IWANN 2019 - Gran Canaria, Spain
Duration: Jun 12 2019Jun 14 2019

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11507 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference15th International Work-Conference on Artificial Neural Networks, IWANN 2019
CountrySpain
CityGran Canaria
Period6/12/196/14/19

Fingerprint

Generative Models
Empirical Study
Neurons
Neuron
Unsupervised Learning
Cancel
Image Reconstruction
Hierarchical Model
Image reconstruction
Statistical property
Statistical Analysis
Statistical methods
Pixel
Pixels
Semantics
Vision

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Zanetti, A., Testolin, A., Zorzi, M., & Wawrzynski, P. (2019). Numerosity Representation in InfoGAN: An Empirical Study. In I. Rojas, G. Joya, & A. Catala (Eds.), Advances in Computational Intelligence - 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Proceedings (pp. 49-60). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11507 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-20518-8_5

Numerosity Representation in InfoGAN : An Empirical Study. / Zanetti, Andrea; Testolin, Alberto; Zorzi, Marco; Wawrzynski, Pawel.

Advances in Computational Intelligence - 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Proceedings. ed. / Ignacio Rojas; Gonzalo Joya; Andreu Catala. Springer Verlag, 2019. p. 49-60 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11507 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Zanetti, A, Testolin, A, Zorzi, M & Wawrzynski, P 2019, Numerosity Representation in InfoGAN: An Empirical Study. in I Rojas, G Joya & A Catala (eds), Advances in Computational Intelligence - 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11507 LNCS, Springer Verlag, pp. 49-60, 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Gran Canaria, Spain, 6/12/19. https://doi.org/10.1007/978-3-030-20518-8_5
Zanetti A, Testolin A, Zorzi M, Wawrzynski P. Numerosity Representation in InfoGAN: An Empirical Study. In Rojas I, Joya G, Catala A, editors, Advances in Computational Intelligence - 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Proceedings. Springer Verlag. 2019. p. 49-60. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-20518-8_5
Zanetti, Andrea ; Testolin, Alberto ; Zorzi, Marco ; Wawrzynski, Pawel. / Numerosity Representation in InfoGAN : An Empirical Study. Advances in Computational Intelligence - 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Proceedings. editor / Ignacio Rojas ; Gonzalo Joya ; Andreu Catala. Springer Verlag, 2019. pp. 49-60 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{c52b2e96ae654a37990805584be8d897,
title = "Numerosity Representation in InfoGAN: An Empirical Study",
abstract = "It has been shown that “visual numerosity emerges as a statistical property of images in ‘deep networks’ that learn a hierarchical generative model of the sensory input”, through unsupervised deep learning [1]. The original deep generative model was based on stochastic neurons and, more importantly, on input (image) reconstruction. Statistical analysis highlighted a correlation between the numerosity present in the input and the population activity of some neurons in the second hidden layer of the network, whereas population activity of neurons in the first hidden layer correlated with total area (i.e., number of pixels) of the objects in the image. Here we further investigate whether numerosity information can be isolated as a disentangled factor of variation of the visual input. We train in unsupervised and semi-supervised fashion a latent-space generative model that has been shown capable of disentangling relevant semantic features in a variety of complex datasets, and we test its generative performance under different conditions. We then propose an approach to the problem based on the assumption that, in order to let numerosity emerge as disentangled factor of variation, we need to cancel out the sources of variation at graphical level.",
author = "Andrea Zanetti and Alberto Testolin and Marco Zorzi and Pawel Wawrzynski",
year = "2019",
month = "1",
day = "1",
doi = "10.1007/978-3-030-20518-8_5",
language = "English",
isbn = "9783030205171",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "49--60",
editor = "Ignacio Rojas and Gonzalo Joya and Andreu Catala",
booktitle = "Advances in Computational Intelligence - 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Proceedings",
address = "Germany",

}

TY - GEN

T1 - Numerosity Representation in InfoGAN

T2 - An Empirical Study

AU - Zanetti, Andrea

AU - Testolin, Alberto

AU - Zorzi, Marco

AU - Wawrzynski, Pawel

PY - 2019/1/1

Y1 - 2019/1/1

N2 - It has been shown that “visual numerosity emerges as a statistical property of images in ‘deep networks’ that learn a hierarchical generative model of the sensory input”, through unsupervised deep learning [1]. The original deep generative model was based on stochastic neurons and, more importantly, on input (image) reconstruction. Statistical analysis highlighted a correlation between the numerosity present in the input and the population activity of some neurons in the second hidden layer of the network, whereas population activity of neurons in the first hidden layer correlated with total area (i.e., number of pixels) of the objects in the image. Here we further investigate whether numerosity information can be isolated as a disentangled factor of variation of the visual input. We train in unsupervised and semi-supervised fashion a latent-space generative model that has been shown capable of disentangling relevant semantic features in a variety of complex datasets, and we test its generative performance under different conditions. We then propose an approach to the problem based on the assumption that, in order to let numerosity emerge as disentangled factor of variation, we need to cancel out the sources of variation at graphical level.

AB - It has been shown that “visual numerosity emerges as a statistical property of images in ‘deep networks’ that learn a hierarchical generative model of the sensory input”, through unsupervised deep learning [1]. The original deep generative model was based on stochastic neurons and, more importantly, on input (image) reconstruction. Statistical analysis highlighted a correlation between the numerosity present in the input and the population activity of some neurons in the second hidden layer of the network, whereas population activity of neurons in the first hidden layer correlated with total area (i.e., number of pixels) of the objects in the image. Here we further investigate whether numerosity information can be isolated as a disentangled factor of variation of the visual input. We train in unsupervised and semi-supervised fashion a latent-space generative model that has been shown capable of disentangling relevant semantic features in a variety of complex datasets, and we test its generative performance under different conditions. We then propose an approach to the problem based on the assumption that, in order to let numerosity emerge as disentangled factor of variation, we need to cancel out the sources of variation at graphical level.

UR - http://www.scopus.com/inward/record.url?scp=85067572075&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85067572075&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-20518-8_5

DO - 10.1007/978-3-030-20518-8_5

M3 - Conference contribution

AN - SCOPUS:85067572075

SN - 9783030205171

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 49

EP - 60

BT - Advances in Computational Intelligence - 15th International Work-Conference on Artificial Neural Networks, IWANN 2019, Proceedings

A2 - Rojas, Ignacio

A2 - Joya, Gonzalo

A2 - Catala, Andreu

PB - Springer Verlag

ER -