Joint optimization of an autoencoder for clustering and embedding

Publikation: Beiträge in ZeitschriftenZeitschriftenaufsätzeForschungbegutachtet

Standard

Joint optimization of an autoencoder for clustering and embedding. / Boubekki, Ahcène; Kampffmeyer, Michael; Brefeld, Ulf et al.
in: Machine Learning, Jahrgang 110, Nr. 7, 01.07.2021, S. 1901-1937.

Publikation: Beiträge in ZeitschriftenZeitschriftenaufsätzeForschungbegutachtet

Harvard

APA

Vancouver

Boubekki A, Kampffmeyer M, Brefeld U, Jenssen R. Joint optimization of an autoencoder for clustering and embedding. Machine Learning. 2021 Jul 1;110(7):1901-1937. doi: 10.1007/s10994-021-06015-5

Bibtex

@article{5d22629599494a47993f95b18dfd860c,
title = "Joint optimization of an autoencoder for clustering and embedding",
abstract = "Deep embedded clustering has become a dominating approach to unsupervised categorization of objects with deep neural networks. The optimization of the most popular methods alternates between the training of a deep autoencoder and a k-means clustering of the autoencoder{\textquoteright}s embedding. The diachronic setting, however, prevents the former to benefit from valuable information acquired by the latter. In this paper, we present an alternative where the autoencoder and the clustering are learned simultaneously. This is achieved by providing novel theoretical insight, where we show that the objective function of a certain class of Gaussian mixture models (GMM{\textquoteright}s) can naturally be rephrased as the loss function of a one-hidden layer autoencoder thus inheriting the built-in clustering capabilities of the GMM. That simple neural network, referred to as the clustering module, can be integrated into a deep autoencoder resulting in a deep clustering model able to jointly learn a clustering and an embedding. Experiments confirm the equivalence between the clustering module and Gaussian mixture models. Further evaluations affirm the empirical relevance of our deep architecture as it outperforms related baselines on several data sets.",
keywords = "Clustering, Deep autoencoders, Embedding, Gaussian mixture models, k-means, Informatics, Business informatics",
author = "Ahc{\`e}ne Boubekki and Michael Kampffmeyer and Ulf Brefeld and Robert Jenssen",
year = "2021",
month = jul,
day = "1",
doi = "10.1007/s10994-021-06015-5",
language = "English",
volume = "110",
pages = "1901--1937",
journal = "Machine Learning",
issn = "0885-6125",
publisher = "Springer US",
number = "7",

}

RIS

TY - JOUR

T1 - Joint optimization of an autoencoder for clustering and embedding

AU - Boubekki, Ahcène

AU - Kampffmeyer, Michael

AU - Brefeld, Ulf

AU - Jenssen, Robert

PY - 2021/7/1

Y1 - 2021/7/1

N2 - Deep embedded clustering has become a dominating approach to unsupervised categorization of objects with deep neural networks. The optimization of the most popular methods alternates between the training of a deep autoencoder and a k-means clustering of the autoencoder’s embedding. The diachronic setting, however, prevents the former to benefit from valuable information acquired by the latter. In this paper, we present an alternative where the autoencoder and the clustering are learned simultaneously. This is achieved by providing novel theoretical insight, where we show that the objective function of a certain class of Gaussian mixture models (GMM’s) can naturally be rephrased as the loss function of a one-hidden layer autoencoder thus inheriting the built-in clustering capabilities of the GMM. That simple neural network, referred to as the clustering module, can be integrated into a deep autoencoder resulting in a deep clustering model able to jointly learn a clustering and an embedding. Experiments confirm the equivalence between the clustering module and Gaussian mixture models. Further evaluations affirm the empirical relevance of our deep architecture as it outperforms related baselines on several data sets.

AB - Deep embedded clustering has become a dominating approach to unsupervised categorization of objects with deep neural networks. The optimization of the most popular methods alternates between the training of a deep autoencoder and a k-means clustering of the autoencoder’s embedding. The diachronic setting, however, prevents the former to benefit from valuable information acquired by the latter. In this paper, we present an alternative where the autoencoder and the clustering are learned simultaneously. This is achieved by providing novel theoretical insight, where we show that the objective function of a certain class of Gaussian mixture models (GMM’s) can naturally be rephrased as the loss function of a one-hidden layer autoencoder thus inheriting the built-in clustering capabilities of the GMM. That simple neural network, referred to as the clustering module, can be integrated into a deep autoencoder resulting in a deep clustering model able to jointly learn a clustering and an embedding. Experiments confirm the equivalence between the clustering module and Gaussian mixture models. Further evaluations affirm the empirical relevance of our deep architecture as it outperforms related baselines on several data sets.

KW - Clustering

KW - Deep autoencoders

KW - Embedding

KW - Gaussian mixture models

KW - k-means

KW - Informatics

KW - Business informatics

UR - http://www.scopus.com/inward/record.url?scp=85109174419&partnerID=8YFLogxK

U2 - 10.1007/s10994-021-06015-5

DO - 10.1007/s10994-021-06015-5

M3 - Journal articles

AN - SCOPUS:85109174419

VL - 110

SP - 1901

EP - 1937

JO - Machine Learning

JF - Machine Learning

SN - 0885-6125

IS - 7

ER -

Dokumente

DOI