Investigating the Corruption Robustness of Image Classifiers with Random Lp-norm Corruptions
This addresses the need for more reliable and safe image classifiers in real-world applications, though it is incremental as it builds on existing data augmentation schemes.
The study tackled the problem of improving image classifier robustness against random p-norm corruptions by augmenting training and test data, finding that combining p-norm corruptions in training significantly enhances robustness beyond existing state-of-the-art methods.
Robustness is a fundamental property of machine learning classifiers required to achieve safety and reliability. In the field of adversarial robustness of image classifiers, robustness is commonly defined as the stability of a model to all input changes within a p-norm distance. However, in the field of random corruption robustness, variations observed in the real world are used, while p-norm corruptions are rarely considered. This study investigates the use of random p-norm corruptions to augment the training and test data of image classifiers. We evaluate the model robustness against imperceptible random p-norm corruptions and propose a novel robustness metric. We empirically investigate whether robustness transfers across different p-norms and derive conclusions on which p-norm corruptions a model should be trained and evaluated. We find that training data augmentation with a combination of p-norm corruptions significantly improves corruption robustness, even on top of state-of-the-art data augmentation schemes.