Variational Inference of Finite Asymmetric Gaussian Mixture Models

Mixture models are a popular unsupervised learning technique useful for discovering homogeneous clusters in unlabeled data. A key research problem lies in the accurate and efficient determination of their associated parameters. Variational inference has recently risen as a prominent parameter learning approach. Hence, in this research, we propose a variational Bayes learning framework for asymmetric Gaussian mixture model. Unlike Gaussian mixture models, these models incorporate the asymmetric shape of data and are adaptive to different conditions in real-word image processing domains. Experimental results show the merit of the proposed approach.