Reviews: Infinite-Horizon Gaussian Processes

This submission proposes to use the stationary distribution of the stochastic function under the GP model for the prediction of streaming data. The required matrices follow the DARE, and the authors propose to solve for these using standard matrix libraries. The method is extended to non-Gaussian likelihoods. Hyper-parameter adjustments based on mini-batches are proposed. Experimental results on four real data sets are given.