Hybrid Generative Models for Two-Dimensional Datasets

Hoda Shajari; Jaemoon Lee; Sanjay Ranka; Anand Rangarajan

doi:10.1007/978-3-030-86340-1_50

Hybrid Generative Models for Two-Dimensional Datasets

Hoda Shajari, Jaemoon Lee, Sanjay Ranka, Anand Rangarajan

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Two-dimensional array-based datasets are pervasive in a variety of domains. Current approaches for generative modeling have typically been limited to conventional image datasets and performed in the pixel domain which does not explicitly capture the correlation between pixels. Additionally, these approaches do not extend to scientific and other applications where each element value is continuous and is not limited to a fixed range. In this paper, we propose a novel approach for generating two-dimensional datasets by moving the computations to the space of representation bases and show its usefulness for two different datasets, one from imaging and another from scientific computing. The proposed approach is general and can be applied to any dataset, representation basis, or generative model. We provide a comprehensive performance comparison of various combinations of generative models and representation basis spaces. We also propose a new evaluation metric which captures the deficiency of generating images in pixel space.

Original language	English
Title of host publication	Artificial Neural Networks and Machine Learning – ICANN 2021 - 30th International Conference on Artificial Neural Networks, Proceedings
Editors	Igor Farkaš, Paolo Masulli, Sebastian Otte, Stefan Wermter
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	623-636
Number of pages	14
ISBN (Print)	9783030863395
DOIs	https://doi.org/10.1007/978-3-030-86340-1_50
State	Published - 2021
Externally published	Yes
Event	30th International Conference on Artificial Neural Networks, ICANN 2021 - Virtual, Online Duration: Sep 14 2021 → Sep 17 2021

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12892 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	30th International Conference on Artificial Neural Networks, ICANN 2021
City	Virtual, Online
Period	09/14/21 → 09/17/21

Funding

Acknowledgments. This material is based upon work supported by the U.S. Department of Energy, Office of Advanced Scientific Computing Research, Scientific Discovery through Advanced Computing (SciDAC) program under Award Number DESC0021320.

Keywords

Generative adversarial networks
Generative models
Image representation bases
Independent component analysis
Normalizing flows

Access to Document

10.1007/978-3-030-86340-1_50

Cite this

Shajari, H., Lee, J., Ranka, S., & Rangarajan, A. (2021). Hybrid Generative Models for Two-Dimensional Datasets. In I. Farkaš, P. Masulli, S. Otte, & S. Wermter (Eds.), Artificial Neural Networks and Machine Learning – ICANN 2021 - 30th International Conference on Artificial Neural Networks, Proceedings (pp. 623-636). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12892 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-030-86340-1_50

Shajari, Hoda ; Lee, Jaemoon ; Ranka, Sanjay et al. / Hybrid Generative Models for Two-Dimensional Datasets. Artificial Neural Networks and Machine Learning – ICANN 2021 - 30th International Conference on Artificial Neural Networks, Proceedings. editor / Igor Farkaš ; Paolo Masulli ; Sebastian Otte ; Stefan Wermter. Springer Science and Business Media Deutschland GmbH, 2021. pp. 623-636 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{ee76a42695024f6eb06716955cbaf0a6,

title = "Hybrid Generative Models for Two-Dimensional Datasets",

abstract = "Two-dimensional array-based datasets are pervasive in a variety of domains. Current approaches for generative modeling have typically been limited to conventional image datasets and performed in the pixel domain which does not explicitly capture the correlation between pixels. Additionally, these approaches do not extend to scientific and other applications where each element value is continuous and is not limited to a fixed range. In this paper, we propose a novel approach for generating two-dimensional datasets by moving the computations to the space of representation bases and show its usefulness for two different datasets, one from imaging and another from scientific computing. The proposed approach is general and can be applied to any dataset, representation basis, or generative model. We provide a comprehensive performance comparison of various combinations of generative models and representation basis spaces. We also propose a new evaluation metric which captures the deficiency of generating images in pixel space.",

keywords = "Generative adversarial networks, Generative models, Image representation bases, Independent component analysis, Normalizing flows",

author = "Hoda Shajari and Jaemoon Lee and Sanjay Ranka and Anand Rangarajan",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; 30th International Conference on Artificial Neural Networks, ICANN 2021 ; Conference date: 14-09-2021 Through 17-09-2021",

year = "2021",

doi = "10.1007/978-3-030-86340-1\_50",

language = "English",

isbn = "9783030863395",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "623--636",

editor = "Igor Farka{\v s} and Paolo Masulli and Sebastian Otte and Stefan Wermter",

booktitle = "Artificial Neural Networks and Machine Learning – ICANN 2021 - 30th International Conference on Artificial Neural Networks, Proceedings",

}

Shajari, H, Lee, J, Ranka, S & Rangarajan, A 2021, Hybrid Generative Models for Two-Dimensional Datasets. in I Farkaš, P Masulli, S Otte & S Wermter (eds), Artificial Neural Networks and Machine Learning – ICANN 2021 - 30th International Conference on Artificial Neural Networks, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12892 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 623-636, 30th International Conference on Artificial Neural Networks, ICANN 2021, Virtual, Online, 09/14/21. https://doi.org/10.1007/978-3-030-86340-1_50

Hybrid Generative Models for Two-Dimensional Datasets. / Shajari, Hoda; Lee, Jaemoon; Ranka, Sanjay et al.
Artificial Neural Networks and Machine Learning – ICANN 2021 - 30th International Conference on Artificial Neural Networks, Proceedings. ed. / Igor Farkaš; Paolo Masulli; Sebastian Otte; Stefan Wermter. Springer Science and Business Media Deutschland GmbH, 2021. p. 623-636 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12892 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Hybrid Generative Models for Two-Dimensional Datasets

AU - Shajari, Hoda

AU - Lee, Jaemoon

AU - Ranka, Sanjay

AU - Rangarajan, Anand

PY - 2021

Y1 - 2021

N2 - Two-dimensional array-based datasets are pervasive in a variety of domains. Current approaches for generative modeling have typically been limited to conventional image datasets and performed in the pixel domain which does not explicitly capture the correlation between pixels. Additionally, these approaches do not extend to scientific and other applications where each element value is continuous and is not limited to a fixed range. In this paper, we propose a novel approach for generating two-dimensional datasets by moving the computations to the space of representation bases and show its usefulness for two different datasets, one from imaging and another from scientific computing. The proposed approach is general and can be applied to any dataset, representation basis, or generative model. We provide a comprehensive performance comparison of various combinations of generative models and representation basis spaces. We also propose a new evaluation metric which captures the deficiency of generating images in pixel space.

AB - Two-dimensional array-based datasets are pervasive in a variety of domains. Current approaches for generative modeling have typically been limited to conventional image datasets and performed in the pixel domain which does not explicitly capture the correlation between pixels. Additionally, these approaches do not extend to scientific and other applications where each element value is continuous and is not limited to a fixed range. In this paper, we propose a novel approach for generating two-dimensional datasets by moving the computations to the space of representation bases and show its usefulness for two different datasets, one from imaging and another from scientific computing. The proposed approach is general and can be applied to any dataset, representation basis, or generative model. We provide a comprehensive performance comparison of various combinations of generative models and representation basis spaces. We also propose a new evaluation metric which captures the deficiency of generating images in pixel space.

KW - Generative adversarial networks

KW - Generative models

KW - Image representation bases

KW - Independent component analysis

KW - Normalizing flows

UR - https://www.scopus.com/pages/publications/85115689910

U2 - 10.1007/978-3-030-86340-1_50

DO - 10.1007/978-3-030-86340-1_50

M3 - Conference contribution

AN - SCOPUS:85115689910

SN - 9783030863395

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

SP - 623

EP - 636

BT - Artificial Neural Networks and Machine Learning – ICANN 2021 - 30th International Conference on Artificial Neural Networks, Proceedings

A2 - Farkaš, Igor

A2 - Masulli, Paolo

A2 - Otte, Sebastian

A2 - Wermter, Stefan

PB - Springer Science and Business Media Deutschland GmbH

T2 - 30th International Conference on Artificial Neural Networks, ICANN 2021

Y2 - 14 September 2021 through 17 September 2021

ER -

Shajari H, Lee J, Ranka S, Rangarajan A. Hybrid Generative Models for Two-Dimensional Datasets. In Farkaš I, Masulli P, Otte S, Wermter S, editors, Artificial Neural Networks and Machine Learning – ICANN 2021 - 30th International Conference on Artificial Neural Networks, Proceedings. Springer Science and Business Media Deutschland GmbH. 2021. p. 623-636. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-86340-1_50