Supplement: SingleModelUncertaintyEstimationvia StochasticDataCentering APPENDIX

For demonstration, let us consider the1D regression example showedinFigure 1andtrain UQmodels under different trainsample sizes(5,10,50 and 200 respectively). The figure illustrates the predicted function and the associated uncertainty estimates (shaded region around thepredictions). ''' model: network trained with anchoring anchors: set of randomly chosen anchors (ideally from train dist.) Foreach case, we showthe negative log-likelihood for the test data obtained using each of the methods. Note, all metrics were computed as an average from20 random trials of0.8 0.2 train-test split.