University of Cambridge > > Microsoft Research Cambridge, public talks > Using transformed domains to sparsify Gaussian Processes

Using transformed domains to sparsify Gaussian Processes

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Microsoft Research Cambridge Talks Admins.

Abstract: One of the main limitations associated to the use of Gaussian Processes (GPs) to model data is their high computational cost, which is O(n^3) for n training samples. In this talk, I will discuss several recently proposed GP models that rely on the use of a `transformed domain` to achieve sparsity and make computation for large data sets affordable. These proposals are comparable in terms of computational cost to previous approaches, such as the celebrated Sparse Pseudo-Input GP (also known as FITC ), but typically offer superior performance. In particular, I will present the Sparse Spectrum GP (SSGP) as a very fast tool to model stationary processes and the Inter-Domain GP (IDGP) as a general framework for inference that can be used both to understand previous sparse GP models and to design new ones. A variational extension of the latter framework that rigorously addresses the overfilling problem will also be discussed.

This talk is part of the Microsoft Research Cambridge, public talks series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.


© 2006-2023, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity