AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
Preprint
Abstract
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.Keywords
All Related Versions
This publication has 10 references indexed in Scilit:
- Random Erasing Data AugmentationProceedings of the AAAI Conference on Artificial Intelligence, 2020
- CutMix: Regularization Strategy to Train Strong Classifiers With Localizable Features2019
- Data Augmentation Using Random Image Cropping and Patching for Deep CNNsIEEE Transactions on Circuits and Systems for Video Technology, 2019
- A Discussion of 'Adversarial Examples Are Not Bugs, They Are Features': Adversarial Example Researchers Need to Expand What is Meant by 'Robustness'Distill, 2019
- MixUp as Locally Linear Out-of-Manifold RegularizationProceedings of the AAAI Conference on Artificial Intelligence, 2019
- Aggregated Residual Transformations for Deep Neural Networks2017
- Improving the Robustness of Deep Neural Networks via Stability Training2016
- Deep Residual Learning for Image Recognition2016
- Posterior calibration and exploratory analysis for natural language processing models2015
- Unbiased look at dataset bias2011