AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
It addresses robustness to unforeseen data shifts in image classification, which is a critical issue for real-world deployment, but the method is incremental as it builds on existing data augmentation techniques.
The paper tackles the problem of deep neural networks' poor performance under distribution shifts between training and test data by proposing AugMix, a simple data processing method that improves robustness and uncertainty estimates for image classifiers, achieving significant gains such as closing the performance gap by more than half in some cases.
Modern deep neural networks can achieve high accuracy when the training distribution and test distribution are identically distributed, but this assumption is frequently violated in practice. When the train and test distributions are mismatched, accuracy can plummet. Currently there are few techniques that improve robustness to unforeseen data shifts encountered during deployment. In this work, we propose a technique to improve the robustness and uncertainty estimates of image classifiers. We propose AugMix, a data processing technique that is simple to implement, adds limited computational overhead, and helps models withstand unforeseen corruptions. AugMix significantly improves robustness and uncertainty measures on challenging image classification benchmarks, closing the gap between previous methods and the best possible performance in some cases by more than half.