Learning joint latent representations based on information maximization

Fei Ye; Adrian Gheorghe Bors

doi:10.1016/j.ins.2021.03.007

Learning joint latent representations based on information maximization

Fei Ye, Adrian Gheorghe Bors

Computer Science

Research output: Contribution to journal › Article › peer-review

Abstract

Learning disentangled and interpretable representations is an important aspect of information understanding. In this paper, we propose a novel deep learning model representing both discrete and continuous latent variable spaces which can be used in either supervised or unsupervised learning. The proposed model is trained using an optimization function employing the mutual information maximization criterion. For the unsupervised learning setting we define a lower bound to the mutual information between the joint distribution of the latent variables corresponding to the real data and those generated by the model. The maximization of this lower bound during the training induces the learning of disentangled and interpretable data representations. Such representations can be used for attribute manipulation and image editing tasks.

Original language	English
Pages (from-to)	216-236
Number of pages	21
Journal	Information Sciences
Volume	567
Early online date	5 Apr 2021
DOIs	https://doi.org/10.1016/j.ins.2021.03.007
Publication status	Published - Aug 2021

Bibliographical note

Access to Document

10.1016/j.ins.2021.03.007

InfoVAEGAN
© 2021 Elsevier Inc. All rights reserved. This is an author-produced version of the published paper. Uploaded in accordance with the publisher’s self-archiving policy.
Accepted author manuscript, 6.44 MBLicence: CC BY-NC-ND

Cite this

@article{55ba76d5997e4467a73feb279f3bc076,

title = "Learning joint latent representations based on information maximization",

abstract = "Learning disentangled and interpretable representations is an important aspect of information understanding. In this paper, we propose a novel deep learning model representing both discrete and continuous latent variable spaces which can be used in either supervised or unsupervised learning. The proposed model is trained using an optimization function employing the mutual information maximization criterion. For the unsupervised learning setting we define a lower bound to the mutual information between the joint distribution of the latent variables corresponding to the real data and those generated by the model. The maximization of this lower bound during the training induces the learning of disentangled and interpretable data representations. Such representations can be used for attribute manipulation and image editing tasks.",

author = "Fei Ye and Bors, {Adrian Gheorghe}",

note = "{\textcopyright} 2021 Elsevier Inc. All rights reserved. This is an author-produced version of the published paper. Uploaded in accordance with the publisher{\textquoteright}s self-archiving policy. ",

year = "2021",

month = aug,

doi = "10.1016/j.ins.2021.03.007",

language = "English",

volume = "567",

pages = "216--236",

journal = "Information Sciences",

issn = "0020-0255",

publisher = "Elsevier",

}

TY - JOUR

T1 - Learning joint latent representations based on information maximization

AU - Ye, Fei

AU - Bors, Adrian Gheorghe

PY - 2021/8

Y1 - 2021/8

N2 - Learning disentangled and interpretable representations is an important aspect of information understanding. In this paper, we propose a novel deep learning model representing both discrete and continuous latent variable spaces which can be used in either supervised or unsupervised learning. The proposed model is trained using an optimization function employing the mutual information maximization criterion. For the unsupervised learning setting we define a lower bound to the mutual information between the joint distribution of the latent variables corresponding to the real data and those generated by the model. The maximization of this lower bound during the training induces the learning of disentangled and interpretable data representations. Such representations can be used for attribute manipulation and image editing tasks.

AB - Learning disentangled and interpretable representations is an important aspect of information understanding. In this paper, we propose a novel deep learning model representing both discrete and continuous latent variable spaces which can be used in either supervised or unsupervised learning. The proposed model is trained using an optimization function employing the mutual information maximization criterion. For the unsupervised learning setting we define a lower bound to the mutual information between the joint distribution of the latent variables corresponding to the real data and those generated by the model. The maximization of this lower bound during the training induces the learning of disentangled and interpretable data representations. Such representations can be used for attribute manipulation and image editing tasks.

U2 - 10.1016/j.ins.2021.03.007

DO - 10.1016/j.ins.2021.03.007

M3 - Article

VL - 567

SP - 216

EP - 236

JO - Information Sciences

JF - Information Sciences

SN - 0020-0255

ER -