Probabilistic blind source separation for data with network structures
Katrin Illner · mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) · 2015
In this thesis we aim to identify meaningful signals from observed multivariate mixtures using available structural information of the data and probabilistic modeling. In a blind source separation (BSS) approach for time series data we investigate the mixing pattern using limiting distributions of the mixing process. We further propose a new BSS method for network data using stationary Bayesian networks. Relevance of the method is illustrated by an application to gene expression data.