Clustering with Deep Learning: Taxonomy and New Methods

By Elie Aljalbout, Vladimir Golkov, Yawar Siddiqui, Daniel Cremers. 2018

265 citations recorded by the source corpus.

Clustering is a fundamental machine learning method. The quality of its results is dependent on the data distribution. For this reason, deep neural networks can be used for learning better representations of the data. In this paper, we propose a systematic taxonomy for clustering with deep learning, in addition to a review of methods from the field. Based on our taxonomy, creating new methods is more straightforward. We also propose a new approach which is built on the taxonomy and surpasses some of the limitations of some previous work. Our experimental evaluation on image datasets shows that the method approaches state-of-the-art clustering quality, and performs better in some cases.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Set2Graph: Learning Graphs From Sets (2020, methodology): “Deep clustering is a large field (Aljalbout et al., 2018) and we restrict our attention to the methods that are most related to ours.”
Deep Self-Organization: Interpretable Discrete Representation Learning on Time Series (2018, methodology): “This includes methods that perform clustering in the latent space of (variational) autoencoders [2] or use a mixture of autoencoders for the clustering [25, 37].”
A Survey of Clustering With Deep Learning: From the Perspective of Network Architecture (2018, background): “Recently, an overview of deep clustering was proposed in [19] to reviewmost remarkable algorithms in this field.”
Deep Amortized Clustering (2019, background): “See Aljalbout et al. (2018) for a comprehensive survey on this line of work.”
Neural Networks and Deep Learning (2018, unknown): “The use of autoencoders for outlier detection is explored in [64, 181, 564], and a survey on the use in clustering is provided in [8].”
Centroid Based Concept Learning for RGB-D Indoor Scene Classification (2019, background): “Combinations of clustering and deep neural networks have been developed in the past [1].”
SOM-VAE: Interpretable Discrete Representation Learning on Time Series (2019, methodology): “This includes methods that perform clustering in the latent space of (variational) autoencoders (Aljalbout et al., 2018) or use a mixture of autoencoders for the clustering (Zhang et al., 2017; Locatello et al., 2018).”
Associative Deep Clustering: Training a Classification Network with No Labels (2018, background): “[43, 17]), are at risk of producing trivial representations, and/or require some labeled data [18] or clustering-unrelated losses (for example additional autoencoder loss [20, 41, 42, 26, 44, 5, 1]) to (pre-)train (parts of) the network.”
Deep Learning for Anomaly Detection (2020, background): “Learning feature representations specifically tailored for a clustering algorithm can well guarantee its performance on different datasets [5].”
Meta-Learning to Cluster (2019, methodology): “Aljalbout et al. [2018] provide an overview of all major frameworks that use deep learning for clustering tasks.”