Learning interpretable disentangled representations using adversarial VAEs

Mhd Hasan Sarhan; Abouzar Eslami; Nassir Navab; Shadi Albarqouni

doi:10.1007/978-3-030-33391-1_5

Learning interpretable disentangled representations using adversarial VAEs

Mhd Hasan Sarhan
, Abouzar Eslami
, Nassir Navab
, Shadi Albarqouni

Informatics 16 - Chair of Computer Aided Medical Procedures

Technical University of Munich
Carl Zeiss
Johns Hopkins University

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

16 Scopus citations

Abstract

Learning Interpretable representation in medical applications is becoming essential for adopting data-driven models into clinical practice. It has been recently shown that learning a disentangled feature representation is important for a more compact and explainable representation of the data. In this paper, we introduce a novel adversarial variational autoencoder with a total correlation constraint to enforce independence on the latent representation while preserving the reconstruction fidelity. Our proposed method is validated on a publicly available dataset showing that the learned disentangled representation is not only interpretable, but also superior to the state-of-the-art methods. We report a relative improvement of 81.50 % in terms of disentanglement, 11.60 % in clustering, and 2 % in supervised classification with a few amount of labeled data.

Original language	English
Title of host publication	Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data First MICCAI Workshop, DART 2019 and First International Workshop, MIL3ID 2019 Shenzhen, Held in Conjunction with MICCAI 2019 Shenzhen, 2019 Proceedings
Editors	Qian Wang, Fausto Milletari, Nicola Rieke, Hien V. Nguyen, Badri Roysam, Shadi Albarqouni, M. Jorge Cardoso, Ziyue Xu, Konstantinos Kamnitsas, Vishal Patel, Steve Jiang, Kevin Zhou, Khoa Luu, Ngan Le
Publisher	Springer
Pages	37-44
Number of pages	8
ISBN (Print)	9783030333904
DOIs	https://doi.org/10.1007/978-3-030-33391-1_5
State	Published - 2019
Event	1st MICCAI Workshop on Domain Adaptation and Representation Transfer, DART 2019, and the 1st International Workshop on Medical Image Learning with Less Labels and Imperfect Data, MIL3ID 2019, held in conjunction with 22nd International Conference on Medical Image Computing and Computer- Assisted Intervention, MICCAI 2019 - Shenzhen, China Duration: 13 Oct 2019 → 17 Oct 2019

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11795 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	1st MICCAI Workshop on Domain Adaptation and Representation Transfer, DART 2019, and the 1st International Workshop on Medical Image Learning with Less Labels and Imperfect Data, MIL3ID 2019, held in conjunction with 22nd International Conference on Medical Image Computing and Computer- Assisted Intervention, MICCAI 2019
Country/Territory	China
City	Shenzhen
Period	13/10/19 → 17/10/19

Keywords

Deep learning
Disentangled representation
Interpretability

Access to Document

10.1007/978-3-030-33391-1_5

Cite this

Sarhan, M. H., Eslami, A., Navab, N., & Albarqouni, S. (2019). Learning interpretable disentangled representations using adversarial VAEs. In Q. Wang, F. Milletari, N. Rieke, H. V. Nguyen, B. Roysam, S. Albarqouni, M. J. Cardoso, Z. Xu, K. Kamnitsas, V. Patel, S. Jiang, K. Zhou, K. Luu, & N. Le (Eds.), Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data First MICCAI Workshop, DART 2019 and First International Workshop, MIL3ID 2019 Shenzhen, Held in Conjunction with MICCAI 2019 Shenzhen, 2019 Proceedings (pp. 37-44). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11795 LNCS). Springer. https://doi.org/10.1007/978-3-030-33391-1_5

Sarhan, Mhd Hasan ; Eslami, Abouzar ; Navab, Nassir et al. / Learning interpretable disentangled representations using adversarial VAEs. Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data First MICCAI Workshop, DART 2019 and First International Workshop, MIL3ID 2019 Shenzhen, Held in Conjunction with MICCAI 2019 Shenzhen, 2019 Proceedings. editor / Qian Wang ; Fausto Milletari ; Nicola Rieke ; Hien V. Nguyen ; Badri Roysam ; Shadi Albarqouni ; M. Jorge Cardoso ; Ziyue Xu ; Konstantinos Kamnitsas ; Vishal Patel ; Steve Jiang ; Kevin Zhou ; Khoa Luu ; Ngan Le. Springer, 2019. pp. 37-44 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{c09d07231452440ba312df7fd855b99d,

title = "Learning interpretable disentangled representations using adversarial VAEs",

abstract = "Learning Interpretable representation in medical applications is becoming essential for adopting data-driven models into clinical practice. It has been recently shown that learning a disentangled feature representation is important for a more compact and explainable representation of the data. In this paper, we introduce a novel adversarial variational autoencoder with a total correlation constraint to enforce independence on the latent representation while preserving the reconstruction fidelity. Our proposed method is validated on a publicly available dataset showing that the learned disentangled representation is not only interpretable, but also superior to the state-of-the-art methods. We report a relative improvement of 81.50 \% in terms of disentanglement, 11.60 \% in clustering, and 2 \% in supervised classification with a few amount of labeled data.",

keywords = "Deep learning, Disentangled representation, Interpretability",

author = "Sarhan, \{Mhd Hasan\} and Abouzar Eslami and Nassir Navab and Shadi Albarqouni",

note = "Publisher Copyright: {\textcopyright} Springer Nature Switzerland AG 2019.; 1st MICCAI Workshop on Domain Adaptation and Representation Transfer, DART 2019, and the 1st International Workshop on Medical Image Learning with Less Labels and Imperfect Data, MIL3ID 2019, held in conjunction with 22nd International Conference on Medical Image Computing and Computer- Assisted Intervention, MICCAI 2019 ; Conference date: 13-10-2019 Through 17-10-2019",

year = "2019",

doi = "10.1007/978-3-030-33391-1\_5",

language = "English",

isbn = "9783030333904",

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

publisher = "Springer",

pages = "37--44",

editor = "Qian Wang and Fausto Milletari and Nicola Rieke and Nguyen, \{Hien V.\} and Badri Roysam and Shadi Albarqouni and Cardoso, \{M. Jorge\} and Ziyue Xu and Konstantinos Kamnitsas and Vishal Patel and Steve Jiang and Kevin Zhou and Khoa Luu and Ngan Le",

booktitle = "Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data First MICCAI Workshop, DART 2019 and First International Workshop, MIL3ID 2019 Shenzhen, Held in Conjunction with MICCAI 2019 Shenzhen, 2019 Proceedings",

}

Sarhan, MH, Eslami, A, Navab, N & Albarqouni, S 2019, Learning interpretable disentangled representations using adversarial VAEs. in Q Wang, F Milletari, N Rieke, HV Nguyen, B Roysam, S Albarqouni, MJ Cardoso, Z Xu, K Kamnitsas, V Patel, S Jiang, K Zhou, K Luu & N Le (eds), Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data First MICCAI Workshop, DART 2019 and First International Workshop, MIL3ID 2019 Shenzhen, Held in Conjunction with MICCAI 2019 Shenzhen, 2019 Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11795 LNCS, Springer, pp. 37-44, 1st MICCAI Workshop on Domain Adaptation and Representation Transfer, DART 2019, and the 1st International Workshop on Medical Image Learning with Less Labels and Imperfect Data, MIL3ID 2019, held in conjunction with 22nd International Conference on Medical Image Computing and Computer- Assisted Intervention, MICCAI 2019, Shenzhen, China, 13/10/19. https://doi.org/10.1007/978-3-030-33391-1_5

Learning interpretable disentangled representations using adversarial VAEs. / Sarhan, Mhd Hasan; Eslami, Abouzar; Navab, Nassir et al.
Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data First MICCAI Workshop, DART 2019 and First International Workshop, MIL3ID 2019 Shenzhen, Held in Conjunction with MICCAI 2019 Shenzhen, 2019 Proceedings. ed. / Qian Wang; Fausto Milletari; Nicola Rieke; Hien V. Nguyen; Badri Roysam; Shadi Albarqouni; M. Jorge Cardoso; Ziyue Xu; Konstantinos Kamnitsas; Vishal Patel; Steve Jiang; Kevin Zhou; Khoa Luu; Ngan Le. Springer, 2019. p. 37-44 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11795 LNCS).

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

TY - GEN

T1 - Learning interpretable disentangled representations using adversarial VAEs

AU - Sarhan, Mhd Hasan

AU - Eslami, Abouzar

AU - Navab, Nassir

AU - Albarqouni, Shadi

N1 - Publisher Copyright: © Springer Nature Switzerland AG 2019.

PY - 2019

Y1 - 2019

N2 - Learning Interpretable representation in medical applications is becoming essential for adopting data-driven models into clinical practice. It has been recently shown that learning a disentangled feature representation is important for a more compact and explainable representation of the data. In this paper, we introduce a novel adversarial variational autoencoder with a total correlation constraint to enforce independence on the latent representation while preserving the reconstruction fidelity. Our proposed method is validated on a publicly available dataset showing that the learned disentangled representation is not only interpretable, but also superior to the state-of-the-art methods. We report a relative improvement of 81.50 % in terms of disentanglement, 11.60 % in clustering, and 2 % in supervised classification with a few amount of labeled data.

AB - Learning Interpretable representation in medical applications is becoming essential for adopting data-driven models into clinical practice. It has been recently shown that learning a disentangled feature representation is important for a more compact and explainable representation of the data. In this paper, we introduce a novel adversarial variational autoencoder with a total correlation constraint to enforce independence on the latent representation while preserving the reconstruction fidelity. Our proposed method is validated on a publicly available dataset showing that the learned disentangled representation is not only interpretable, but also superior to the state-of-the-art methods. We report a relative improvement of 81.50 % in terms of disentanglement, 11.60 % in clustering, and 2 % in supervised classification with a few amount of labeled data.

KW - Deep learning

KW - Disentangled representation

KW - Interpretability

UR - https://www.scopus.com/pages/publications/85075654373

U2 - 10.1007/978-3-030-33391-1_5

DO - 10.1007/978-3-030-33391-1_5

M3 - Conference contribution

AN - SCOPUS:85075654373

SN - 9783030333904

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

SP - 37

EP - 44

BT - Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data First MICCAI Workshop, DART 2019 and First International Workshop, MIL3ID 2019 Shenzhen, Held in Conjunction with MICCAI 2019 Shenzhen, 2019 Proceedings

A2 - Wang, Qian

A2 - Milletari, Fausto

A2 - Rieke, Nicola

A2 - Nguyen, Hien V.

A2 - Roysam, Badri

A2 - Albarqouni, Shadi

A2 - Cardoso, M. Jorge

A2 - Xu, Ziyue

A2 - Kamnitsas, Konstantinos

A2 - Patel, Vishal

A2 - Jiang, Steve

A2 - Zhou, Kevin

A2 - Luu, Khoa

A2 - Le, Ngan

PB - Springer

T2 - 1st MICCAI Workshop on Domain Adaptation and Representation Transfer, DART 2019, and the 1st International Workshop on Medical Image Learning with Less Labels and Imperfect Data, MIL3ID 2019, held in conjunction with 22nd International Conference on Medical Image Computing and Computer- Assisted Intervention, MICCAI 2019

Y2 - 13 October 2019 through 17 October 2019

ER -

Sarhan MH, Eslami A, Navab N, Albarqouni S. Learning interpretable disentangled representations using adversarial VAEs. In Wang Q, Milletari F, Rieke N, Nguyen HV, Roysam B, Albarqouni S, Cardoso MJ, Xu Z, Kamnitsas K, Patel V, Jiang S, Zhou K, Luu K, Le N, editors, Domain Adaptation and Representation Transfer and Medical Image Learning with Less Labels and Imperfect Data First MICCAI Workshop, DART 2019 and First International Workshop, MIL3ID 2019 Shenzhen, Held in Conjunction with MICCAI 2019 Shenzhen, 2019 Proceedings. Springer. 2019. p. 37-44. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-33391-1_5

Learning interpretable disentangled representations using adversarial VAEs

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this