Deep clustering with a Dynamic Autoencoder: From reconstruction towards centroids construction

• DOI: 10.1016/j.neunet.2020.07.005
• Publication Date: 2020-07-11T02:36:39Z
• Authors: Nairouz Mrabah; Naimul Mefraz Khan; Riadh Ksantini; Zied Lachiri
• Keywords: FOS: Computer and information sciences; 0301 basic medicine; Computer Science - Machine Learning; Artificial intelligence; Semi-Supervised Learning; Unsupervised Learning; Evolutionary biology; Machine Learning (stat.ML); Pattern recognition (psychology); Unsupervised learning; Pattern Recognition, Automated; Machine Learning (cs.LG); Anomaly Detection in High-Dimensional Data; 03 medical and health sciences; Deep Learning; Cluster analysis; Clustering Analysis; Statistics - Machine Learning; Artificial Intelligence; Meta-Learning; Machine learning; Cluster Analysis; Data mining; Biology; Geography; Deep learning; Autoencoder; Centroid; Computer science; Process (computing); Operating system; Advances in Transfer Learning and Domain Adaptation; Function (biology); Human Action Recognition and Pose Estimation; Computer Science; Physical Sciences; Computer Vision and Pattern Recognition; Benchmark (surveying); Algorithms; Geodesy

<p>In unsupervised learning, there is no apparent straightforward cost function that can capture the significant factors of variations and similarities. Since natural systems have smooth dynamics, an opportunity is lost if an unsupervised objective function remains static during the training process. The absence of concrete supervision suggests that smooth dynamics should be integrated. Compared to classical static cost functions, dynamic objective functions allow to better make use of the gradual and uncertain knowledge acquired through pseudo-supervision. In this paper, we propose Dynamic Autoencoder (DynAE), a novel model for deep clustering that overcomes a clustering-reconstruction trade-off, by gradually and smoothly eliminating the reconstruction objective function in favor of a construction one. Experimental evaluations on benchmark datasets show that our approach achieves state-of-the-art results compared to the most relevant deep clustering methods.</p>