On Wednesday, March 24th, James Ng presented recent work related to nonparametric statistical models based on Gaussian processes on the unit hypersphere. An abstract for James’ talk is below.
Abstract
We study nonparametric statistical models based on Gaussian processes on the unit hypersphere. We derive convergence rates of posterior distributions for various choices of covariance kernels. We then consider the logistic Gaussian process prior for density estimation on the unit hypersphere and propose a variational inference with spherical harmonics features.