An Exponential Family Extension to Principal Component Analysis

Abstract

In this paper, we present a unified probabilistic model for constrained factorisation models, which employs exponential family distributions to represent the constrained factors. Our main objective is to provide a versatile framework, on which prototype models with various constraints can be implemented effortlessly. For learning the proposed stochastic model, Gibbs sampling is employed for model inference. We also demonstrate the utility and versatility of the model by experiments.