Frustratingly Easy Uncertainty Estimation for Distribution Shift
This provides a simple and effective solution for uncertainty estimation in image classification under distribution shift, which is incremental as it builds on existing methods with a straightforward approach.
The paper tackles uncertainty estimation under distribution shift in deep image classification by showing that it can be achieved simply by exposing the model to corrupted images and performing statistical calibration, resulting in superior performance across various shifts and domain adaptation tasks.
Distribution shift is an important concern in deep image classification, produced either by corruption of the source images, or a complete change, with the solution involving domain adaptation. While the primary goal is to improve accuracy under distribution shift, an important secondary goal is uncertainty estimation: evaluating the probability that the prediction of a model is correct. While improving accuracy is hard, uncertainty estimation turns out to be frustratingly easy. Prior works have appended uncertainty estimation into the model and training paradigm in various ways. Instead, we show that we can estimate uncertainty by simply exposing the original model to corrupted images, and performing simple statistical calibration on the image outputs. Our frustratingly easy methods demonstrate superior performance on a wide range of distribution shifts as well as on unsupervised domain adaptation tasks, measured through extensive experimentation.