Parallel Gaussian process with kernel approximation in CUDA

Abstract

This paper introduces a parallel implementation in CUDA/C++ of the Gaussian process with a decomposed kernel. This recent formulation, introduced by Joukov and Kuli\'c (2022), is characterized by an approximated -- but much smaller -- matrix to be inverted compared to plain Gaussian process. However, it exhibits a limitation when dealing with higher-dimensional samples which degrades execution times. The solution presented in this paper relies on parallelizing the computation of the predictive posterior statistics on a GPU using CUDA and its libraries. The CPU code and GPU code are then benchmarked on different CPU-GPU configurations to show the benefits of the parallel implementation on GPU over the CPU.