Uncertainty Estimation Using a Single Deep Deterministic Neural Network

About

We propose a method for training a deterministic deep model that can find and reject out of distribution data points at test time with a single forward pass. Our approach, deterministic uncertainty quantification (DUQ), builds upon ideas of RBF networks. We scale training in these with a novel loss function and centroid updating scheme and match the accuracy of softmax models. By enforcing detectability of changes in the input using a gradient penalty, we are able to reliably detect out of distribution data. Our uncertainty quantification scales well to large datasets, and using a single model, we improve upon or match Deep Ensembles in out of distribution detection on notable difficult dataset pairs such as FashionMNIST vs. MNIST, and CIFAR-10 vs. SVHN.

Joost van Amersfoort, Lewis Smith, Yee Whye Teh, Yarin Gal · 2020

Related benchmarks

Task	Dataset	Result
Image Classification	CIFAR-10 (test)	Accuracy—	3,381
Image Classification	CIFAR-10 (test)	Accuracy89.33	1,063
Image Classification	CIFAR-10	Accuracy94.1	507
Image Classification	FashionMNIST (test)	Accuracy92.4	461
Image Classification	CIFAR-10-C	Accuracy69.01	183
Out-of-Distribution Detection	CIFAR-10 vs SVHN (test)	AUROC0.927	146
Out-of-Distribution Detection	CIFAR-10	AUROC—	121
Out-of-Distribution Detection	CIFAR-100 SVHN in-distribution out-of-distribution (test)	AUROC88.7	111
One-vs-all classification	CIFAR-10	Accuracy71	90
Out-of-Distribution Detection	ImageNet-O	AUROC0.714	74