Multi-Relational Learning with Gaussian Processes
Gaussian process models are routinely used to solve hard machine learning problems. They are attractive because of their flexible non-parametric nature and the predictive distributions they provide for test instances. While past Gaussian process models focused on modeling at most a single relation, we present a generalized GP model, called multi-relational Gaussian process model, that is able to deal with an arbitrary number of relations in a domain of interest. We provide an analysis of our model for several types of relations such as bipartite, directed, and undirected univariate ones. Experimental results on real world datasets verify the usefulness of our approach: exploiting the correlations among different entity types and relations can indeed improve the quality of predictions.