Supplementary Material: Posterior and Computational Uncertainty in Gaussian Processes Jonathan Wenger

This supplementary material contains additional results and in particular proofs for all theoretical statements. Then Algorithm 1 recovers the pivoted Cholesky decomposition, i.e. it holds for all It holds by assumption and eq. S1.3 Conjugate Gradient MethodAlgorithm S3: Preconditioned Conjugate Gradient Method [38] Input: kernel matrix ˆ K, labels y, prior mean µ, preconditioner ˆ P Output: representer weights v We prove the claim by induction. By the form of preconditioned deflated CG given in Algorithm 3.6 of Saad et al. This proves the first claim. We prove the claims by induction.