Kernel Interpolation with Sparse Grids
This addresses a computational bottleneck for Gaussian process inference in high-dimensional settings, though it appears incremental as an extension of the SKI framework.
The paper tackles the exponential scaling problem of structured kernel interpolation (SKI) for Gaussian processes in high dimensions by proposing sparse grids, which reduce grid size growth with dimension while maintaining accuracy. They demonstrate this approach scales SKI to higher dimensions with a novel nearly linear-time matrix-vector multiplication algorithm.
Structured kernel interpolation (SKI) accelerates Gaussian process (GP) inference by interpolating the kernel covariance function using a dense grid of inducing points, whose corresponding kernel matrix is highly structured and thus amenable to fast linear algebra. Unfortunately, SKI scales poorly in the dimension of the input points, since the dense grid size grows exponentially with the dimension. To mitigate this issue, we propose the use of sparse grids within the SKI framework. These grids enable accurate interpolation, but with a number of points growing more slowly with dimension. We contribute a novel nearly linear time matrix-vector multiplication algorithm for the sparse grid kernel matrix. Next, we describe how sparse grids can be combined with an efficient interpolation scheme based on simplices. With these changes, we demonstrate that SKI can be scaled to higher dimensions while maintaining accuracy.