A Generative Clustering Ensemble Model and Its Application in IoT Data Analysis

Figure 1

Overall framework of the proposed GCE model. The proposed model consists of three components. (i) In the structure extraction component, structure information among data samples is extracted from a series of base clusterings at first. (ii) In the encoder component, data characteristics and structure information are encoded integrated by a GCN module, which learns parameters for the distribution of latent representation. (iii) In the decoder component, is reconstructed by a generator , and the representation sampled from the learned latent distribution is assumed to lie in a cluster characterized by a Gaussian component of the GMM. The objective of representation learning and clustering ensemble are jointly optimized by maximizing the ELBO of GCE, which is calculated and backpropagated to the latent representation.