ITE Technical Group Submission System
Conference Paper's Information
Online Proceedings
[Sign in]
 Go Top Page Go Previous   [Japanese] / [English] 

Paper Abstract and Keywords
Presentation 2022-02-21 16:45
A Note on Disentanglement Using Deep Generative Model Based on Variational Autoencoder -- Introduction of Regularization Losses Based on Metrics of Disentangled Representation --
Nao Nakagawa, Ren Togo, Takahiro Ogawa, Miki Haseyama (Hokkaido Univ.)
Abstract (in Japanese) (See Japanese page) 
(in English) In this paper, we study disentangled representation learning using a deep generative model based on Variational Autoencoder (VAE). The goal of disentanglement is to obtain a latent representation in which single latent variables correspond to single factors of variation. Although several unsupervised methods have been proposed for disentanglement by imposing the element-wise independence of latent variables, it has been shown that independence does not guarantee disentanglement. Hence, we propose a novel disentanglement method using a VAE-based model whose loss function includes a regularization loss based on a differentiable disentanglement metric. Our method disentangles the representation by applying gradient descent directly to a disentanglement metric function. We first validate the behavior of the various disentanglement metrics and then show the effectiveness of our method.
Keyword (in Japanese) (See Japanese page) 
(in English) deep learning / representation learning / deep generative model / disentanglement / variational autoencoder / / /  
Reference Info. ITE Tech. Rep., vol. 46, no. 6, ME2022-44, pp. 97-102, Feb. 2022.
Paper # ME2022-44 
Date of Issue 2022-02-14 (MMS, ME, AIT) 
ISSN Print edition: ISSN 1342-6893    Online edition: ISSN 2424-1970
Download PDF

Conference Information
Conference Date 2022-02-21 - 2022-02-22 
Place (in Japanese) (See Japanese page) 
Place (in English) online 
Topics (in Japanese) (See Japanese page) 
Topics (in English)  
Paper Information
Registration To ME 
Conference Code 2022-02-AIT-ME-MMS-IE-ITS 
Language Japanese 
Title (in Japanese) (See Japanese page) 
Sub Title (in Japanese) (See Japanese page) 
Title (in English) A Note on Disentanglement Using Deep Generative Model Based on Variational Autoencoder 
Sub Title (in English) Introduction of Regularization Losses Based on Metrics of Disentangled Representation 
Keyword(1) deep learning  
Keyword(2) representation learning  
Keyword(3) deep generative model  
Keyword(4) disentanglement  
Keyword(5) variational autoencoder  
1st Author's Name Nao Nakagawa  
1st Author's Affiliation Hokkaido University (Hokkaido Univ.)
2nd Author's Name Ren Togo  
2nd Author's Affiliation Hokkaido University (Hokkaido Univ.)
3rd Author's Name Takahiro Ogawa  
3rd Author's Affiliation Hokkaido University (Hokkaido Univ.)
4th Author's Name Miki Haseyama  
4th Author's Affiliation Hokkaido University (Hokkaido Univ.)
5th Author's Name  
5th Author's Affiliation ()
6th Author's Name  
6th Author's Affiliation ()
7th Author's Name  
7th Author's Affiliation ()
8th Author's Name  
8th Author's Affiliation ()
9th Author's Name  
9th Author's Affiliation ()
10th Author's Name  
10th Author's Affiliation ()
11th Author's Name  
11th Author's Affiliation ()
12th Author's Name  
12th Author's Affiliation ()
13th Author's Name  
13th Author's Affiliation ()
14th Author's Name  
14th Author's Affiliation ()
15th Author's Name  
15th Author's Affiliation ()
16th Author's Name  
16th Author's Affiliation ()
17th Author's Name  
17th Author's Affiliation ()
18th Author's Name  
18th Author's Affiliation ()
19th Author's Name  
19th Author's Affiliation ()
20th Author's Name  
20th Author's Affiliation ()
Speaker Author-1 
Date Time 2022-02-21 16:45:00 
Presentation Time 15 minutes 
Registration for ME 
Paper # MMS2022-19, ME2022-44, AIT2022-19 
Volume (vol) vol.46 
Number (no) no.6 
Page pp.97-102 
Date of Issue 2022-02-14 (MMS, ME, AIT) 

[Return to Top Page]

[Return to ITE Web Page]

The Institute of Image Information and Television Engineers (ITE), Japan