Semi-Supervised Learning for Stratified Networks

This research is concerned with developing a semi-supervised learning algorithm for stratified networks. In stratified networks, labels in one stratum can benefit predictions in other strata through inter-stratum connections so dealing with inter-stratum connections is important. Technically, the problem of non-squareness and sparseness involved in matrix inversion for interstratum connections must be solved. In order to verify the validity of the algorithm, it was applied on disease-symptom network structure to predict cooccurrence of two diseases.