Table 1 present the evaluation results. We also evaluate our method on the CIFAR benchmark with various model architectures. The best three results are highlighted with red, blue, and cyan. The best two results are highlighted with red and blue . Table 2 compares the performance against all the post hoc baselines.

We remark that the Dirichlet distribution with a uniform parameter of one can be easily approximated by independent exponential distribution. We use this block bootstrap as a default of the NeuBoots in sequel. We shall rigorously investigate asymptotic equivalence between the blocked bootstrap and the non-blocked bootstrap. To ease the explanation for theory, we introduce some notation here. Recall that the notation is introduced in Section 2 .[

Bootstrapping has been a primary tool for ensemble and uncertainty quantification in machine learning and statistics.