Factorization and Parallelization Clause Samples

Factorization and Parallelization. ‌ To speed up the computation of Bayesian GPLVM even further (on top of the sparse GP approximation), we can employ parallel computation of tasks in modern multi core computational units. In this section, we will have a look at how the Bayesian GPLVM bound can be factorized in a simple view. The factorization of the Bound across dimensions D has been implemented during the course of this thesis and is available through GPy (Sec. 3). We can then take advantage of that and distribute the calculations of partial sums on different cores and collect the sums in a master node. The bound described in this section has useful factorization properties in both, N and D. Here, we will describe in short how these factorizations come to be and can be used to compute in parallel to speed up computation.