Pitfalls of In-Domain Uncertainty Estimation and Ensembling in Deep Learning
This work addresses pitfalls in uncertainty estimation for image classification, which is important for improving reliability in AI applications, but it is incremental as it builds on existing ensembling methods.
The paper tackles the problem of in-domain uncertainty estimation in deep learning for image classification by identifying pitfalls in existing metrics and conducting a broad study of ensembling techniques, showing that many sophisticated methods are equivalent to an ensemble of few independently trained networks in terms of test performance.
Uncertainty estimation and ensembling methods go hand-in-hand. Uncertainty estimation is one of the main benchmarks for assessment of ensembling performance. At the same time, deep learning ensembles have provided state-of-the-art results in uncertainty estimation. In this work, we focus on in-domain uncertainty for image classification. We explore the standards for its quantification and point out pitfalls of existing metrics. Avoiding these pitfalls, we perform a broad study of different ensembling techniques. To provide more insight in this study, we introduce the deep ensemble equivalent score (DEE) and show that many sophisticated ensembling techniques are equivalent to an ensemble of only few independently trained networks in terms of test performance.