Duality model-based approaches to clustering

Kernel spectral clustering models have been successful, being related to least squares support vector machine models that possess primal and dual model representations. It enables to do out-of-sample extensions, work with representative subsets and develop large scale algorithms. Recently, further extensions have been on generative kernel principal component analysis through restricted kernel machines and conjugate feature duality. The aim of the research is to explore such new model based approaches and generative models for clustering problems. These will be studied both in primal forms related to (deep) neural networks and dual forms related to kernel-based representations.