What classifiers know what they don't?

TL;DR

This paper introduces UIMNET, a large-scale benchmark for evaluating uncertainty estimates in deep image classifiers, addressing the gap in realistic, high-scale testing for out-of-distribution detection.

Contribution

The paper presents UIMNET, a comprehensive, open-source benchmark with implementations of multiple algorithms and metrics for realistic evaluation of uncertainty in image classification.

Findings

Ensembles of ERM classifiers perform best for uncertainty estimation.

Single MIMO classifiers are competitive alternatives.

UIMNET facilitates reproducible and extensible research in uncertainty estimation.

Abstract

Being uncertain when facing the unknown is key to intelligent decision making. However, machine learning algorithms lack reliable estimates about their predictive uncertainty. This leads to wrong and overly-confident decisions when encountering classes unseen during training. Despite the importance of equipping classifiers with uncertainty estimates ready for the real world, prior work has focused on small datasets and little or no class discrepancy between training and testing data. To close this gap, we introduce UIMNET: a realistic, ImageNet-scale test-bed to evaluate predictive uncertainty estimates for deep image classifiers. Our benchmark provides implementations of eight state-of-the-art algorithms, six uncertainty measures, four in-domain metrics, three out-domain metrics, and a fully automated pipeline to train, calibrate, ensemble, select, and evaluate models. Our test-bed is open-source and all of our results are reproducible from a fixed commit in our repository. Adding new datasets, algorithms, measures, or metrics is a matter of a few lines of code-in so hoping that UIMNET becomes a stepping stone towards realistic, rigorous, and reproducible research in uncertainty estimation. Our results show that ensembles of ERM classifiers as well as single MIMO classifiers are the two best alternatives currently available to measure uncertainty about both in-domain and out-domain classes.