Tucker Gaussian Process for Regression and Collaborative Filtering

We introduce the Tucker Gaussian Process (TGP), an approach to scalable GP learning based on low-rank tensor decompositions. We show that our model is applicable to general regression problems, and is particularly well-suited to grid-structured data and problems where the dependence on covariates is close to being separable. Furthermore, when applied to collaborative filtering, our model provides an effective GP based method that has a low-rank matrix factorisation at its core, and gives a natural and elegant method for incorporating side information.