PROJECT TITLE :

Clustering Analysis via Deep Generative Models With Mixture Models

ABSTRACT:

Clustering is a fundamental problem that crops up frequently in many fields, including pattern recognition, data mining, and machine learning, amongst others. Traditional clustering algorithms, on the other hand, have shallow structures and are unable to excavate the interdependence of complex data features in latent space. This is despite the fact that numerous clustering algorithms have been developed in the past. Recently, deep generative models like autoencoder (AE), variational autoencoder (VAE), and generative adversarial network (GAN) have achieved remarkable success in many unsupervised applications thanks to their capabilities for learning promising latent representations from original data. This success has been made possible by the fact that these models can learn from unsupervised data. In this work, we begin by presenting a novel approach to clustering that is based on both the Wasserstein Generalized Additive Model with Gradient Penalty (WGAN-GP) and the Variance Accumulator Estimator with a Gaussian Mixture Prior. The generator of the WGAN-GP is formulated by drawing samples from the probabilistic decoder of the VAE. This is accomplished by combining the WGAN-GP with the VAE. In addition, a variant of the proposed deep generative model that is based on a Student's-t mixture prior has been developed in order to provide more robust clustering and generation performance when outliers are encountered in the data. This has been accomplished. Experiments on both clustering analysis and the generation of samples are used to demonstrate that our deep generative models are effective. The proposed method can provide more stable training of the model, improve the accuracy of clustering, and generate more realistic samples when compared to other state-of-the-art clustering approaches that are based on deep generative models.