Predictive Classification and Bayesian Inference

Abstract

A general inductive probabilistic framework for clustering and classification is introduced using the principles of Bayesian predictive inference, such that all quantities are jointly modelled and the uncertainty is fully acknowledged through the posterior predictive distribution. Several learning rules have been considered and the theoretical results are extended to acknowledge complex dependencies within the datasets. Multiple probabilistic models have been developed for analysing data from a wide variate of fields of application. State-of-art algorithms are introduced and developed for the model optimization.