Research Blog

Efficiently sampling functions from Gaussian process posteriors

Efficient sampling from Gaussian process posteriors is relevant in practical applications. With Matheron’s rule we decouple the posterior, which allows us to sample functions from the Gaussian process posterior in linear time.

Healing Products of Gaussian Process Experts

Products of Gaussian process experts commonly suffer from poor performance when experts are weak. We propose aggregations and weighting approaches to heal these expert models.

Matérn Gaussian processes on Riemannian manifolds

Gaussian processes are a useful technique for modeling unknown functions. They are used in many application areas, particularly in cases where quantifying uncertainty is important, such as in strategic decision-making systems. We study how to extend this model class to model functions whose domain is a Riemannian manifold, for example, a sphere or cylinder. We do so in a manner which is (a) mathematically well-posed, and (b) constructive enough to allow the kernel to be computed, thereby allowing said processes to be trained with standard methods.