Efficient Gaussian process learning via subspace projections
This work addresses computational bottlenecks in GP learning for moderately large datasets, offering a more efficient method.
The authors tackled the computational inefficiency of Gaussian process (GP) training by introducing a novel projected likelihood objective using lower-dimensional linear projections, achieving superior accuracy and computational efficiency compared to exact GP training and variational sparse GPs across various optimizers, kernels, and moderately large datasets.
We propose a novel training objective for GPs constructed using lower-dimensional linear projections of the data, referred to as \emph{projected likelihood} (PL). We provide a closed-form expression for the information loss related to the PL and empirically show that it can be reduced with random projections on the unit sphere. We show the superiority of the PL, in terms of accuracy and computational efficiency, over the exact GP training and the variational free energy approach to sparse GPs over different optimisers, kernels and datasets of moderately large sizes.