Iterative autoencoding and clustering for unsupervised feature representation

Songlin Du, Takeshi Ikenaga

研究成果: Conference contribution

抄録

Unsupervised feature representation is a challenging problem in machine learning and computer vision. Since manual labels are unavailable for training, it is difficult to reduce the gap between learned features and image semantics. This paper proposes an iterative autoencoding and clustering approach, which consists of an autoencoding sub-network and a classification sub-network, for unsupervised feature representation. On one hand, the autoencoding sub-network maps images to features. On the other hand, using the features generated by the autoencoding sub-network, the classification sub-network maps the features to classes and estimates pseudo labels by clustering the features simultaneously. Through iterations between the feature representation and the pseudo-labels-supervised classification, the gap between features and image semantics is reduced. Experimental results on handwritten digits recognition and objects classification prove that the proposed approach achieves state-of-the-art performance compared with existing methods.

元の言語English
ホスト出版物のタイトル2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019 - Proceedings
出版者Institute of Electrical and Electronics Engineers Inc.
ISBN(電子版)9781728103976
DOI
出版物ステータスPublished - 2019 1 1
イベント2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019 - Sapporo, Japan
継続期間: 2019 5 262019 5 29

出版物シリーズ

名前Proceedings - IEEE International Symposium on Circuits and Systems
2019-May
ISSN(印刷物)0271-4310

Conference

Conference2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019
Japan
Sapporo
期間19/5/2619/5/29

Fingerprint

Labels
Semantics
Computer vision
Learning systems

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

これを引用

Du, S., & Ikenaga, T. (2019). Iterative autoencoding and clustering for unsupervised feature representation. : 2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019 - Proceedings [8702659] (Proceedings - IEEE International Symposium on Circuits and Systems; 巻数 2019-May). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ISCAS.2019.8702659

Iterative autoencoding and clustering for unsupervised feature representation. / Du, Songlin; Ikenaga, Takeshi.

2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2019. 8702659 (Proceedings - IEEE International Symposium on Circuits and Systems; 巻 2019-May).

研究成果: Conference contribution

Du, S & Ikenaga, T 2019, Iterative autoencoding and clustering for unsupervised feature representation. : 2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019 - Proceedings., 8702659, Proceedings - IEEE International Symposium on Circuits and Systems, 巻. 2019-May, Institute of Electrical and Electronics Engineers Inc., 2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019, Sapporo, Japan, 19/5/26. https://doi.org/10.1109/ISCAS.2019.8702659
Du S, Ikenaga T. Iterative autoencoding and clustering for unsupervised feature representation. : 2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019 - Proceedings. Institute of Electrical and Electronics Engineers Inc. 2019. 8702659. (Proceedings - IEEE International Symposium on Circuits and Systems). https://doi.org/10.1109/ISCAS.2019.8702659
Du, Songlin ; Ikenaga, Takeshi. / Iterative autoencoding and clustering for unsupervised feature representation. 2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2019. (Proceedings - IEEE International Symposium on Circuits and Systems).
@inproceedings{3425ce69f35a4f0d8b278e10be574ed0,
title = "Iterative autoencoding and clustering for unsupervised feature representation",
abstract = "Unsupervised feature representation is a challenging problem in machine learning and computer vision. Since manual labels are unavailable for training, it is difficult to reduce the gap between learned features and image semantics. This paper proposes an iterative autoencoding and clustering approach, which consists of an autoencoding sub-network and a classification sub-network, for unsupervised feature representation. On one hand, the autoencoding sub-network maps images to features. On the other hand, using the features generated by the autoencoding sub-network, the classification sub-network maps the features to classes and estimates pseudo labels by clustering the features simultaneously. Through iterations between the feature representation and the pseudo-labels-supervised classification, the gap between features and image semantics is reduced. Experimental results on handwritten digits recognition and objects classification prove that the proposed approach achieves state-of-the-art performance compared with existing methods.",
author = "Songlin Du and Takeshi Ikenaga",
year = "2019",
month = "1",
day = "1",
doi = "10.1109/ISCAS.2019.8702659",
language = "English",
series = "Proceedings - IEEE International Symposium on Circuits and Systems",
publisher = "Institute of Electrical and Electronics Engineers Inc.",
booktitle = "2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019 - Proceedings",

}

TY - GEN

T1 - Iterative autoencoding and clustering for unsupervised feature representation

AU - Du, Songlin

AU - Ikenaga, Takeshi

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Unsupervised feature representation is a challenging problem in machine learning and computer vision. Since manual labels are unavailable for training, it is difficult to reduce the gap between learned features and image semantics. This paper proposes an iterative autoencoding and clustering approach, which consists of an autoencoding sub-network and a classification sub-network, for unsupervised feature representation. On one hand, the autoencoding sub-network maps images to features. On the other hand, using the features generated by the autoencoding sub-network, the classification sub-network maps the features to classes and estimates pseudo labels by clustering the features simultaneously. Through iterations between the feature representation and the pseudo-labels-supervised classification, the gap between features and image semantics is reduced. Experimental results on handwritten digits recognition and objects classification prove that the proposed approach achieves state-of-the-art performance compared with existing methods.

AB - Unsupervised feature representation is a challenging problem in machine learning and computer vision. Since manual labels are unavailable for training, it is difficult to reduce the gap between learned features and image semantics. This paper proposes an iterative autoencoding and clustering approach, which consists of an autoencoding sub-network and a classification sub-network, for unsupervised feature representation. On one hand, the autoencoding sub-network maps images to features. On the other hand, using the features generated by the autoencoding sub-network, the classification sub-network maps the features to classes and estimates pseudo labels by clustering the features simultaneously. Through iterations between the feature representation and the pseudo-labels-supervised classification, the gap between features and image semantics is reduced. Experimental results on handwritten digits recognition and objects classification prove that the proposed approach achieves state-of-the-art performance compared with existing methods.

UR - http://www.scopus.com/inward/record.url?scp=85066793591&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85066793591&partnerID=8YFLogxK

U2 - 10.1109/ISCAS.2019.8702659

DO - 10.1109/ISCAS.2019.8702659

M3 - Conference contribution

AN - SCOPUS:85066793591

T3 - Proceedings - IEEE International Symposium on Circuits and Systems

BT - 2019 IEEE International Symposium on Circuits and Systems, ISCAS 2019 - Proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

ER -