NeurIPS_wappendix

A.1 Gaussian Process The generative model of a GP is given as f | X N (0,K The parameter of interest is f, and the hyperparameters are nuisance parameters, 2. The parameters of interest are both f and the hyperparameters, 3. The following posteriors are all averaged across the 10 experiments for a particular data size. Figure 6, the five posteriors that gave a multimodal posterior using the RBF kernel are shown. The data sizes are for Gramacy1d: 70, Higdon: 40, Gramacy2d: 40, Branin: 40, and Motorcycle: 10. Note that if the AUC is computed by a metric that is not lower bounded, e.g., the negative marginal The motorcycle simulator is created by fitting a variational GP [Hensman et al., 2015] to the The simulator is fully specified by the following mean standard deviation. The behavior of the two runs are consistent with the other runs.