Improving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation

By Raphael Gontijo Lopes, Dong Yin, Ben Poole, Justin Gilmer, Ekin Dogus Cubuk. 2019

229 citations recorded by the source corpus.

Deploying machine learning systems in the real world requires both high accuracy on clean data and robustness to naturally occurring corruptions. While architectural advances have led to improved accuracy, building robust models remains challenging. Prior work has argued that there is an inherent trade-off between robustness and accuracy, which is exemplified by standard data augment techniques such as Cutout, which improves clean accuracy but not robustness, and additive Gaussian noise, which improves robustness but hurts accuracy. To overcome this trade-off, we introduce Patch Gaussian, a simple augmentation scheme that adds noise to randomly selected patches in an input image. Models trained with Patch Gaussian achieve state of the art on the CIFAR-10 and ImageNetCommon Corruptions benchmarks while also improving accuracy on clean data. We find that this augmentation leads to reduced sensitivity to high frequency noise(similar to Gaussian) while retaining the ability to take advantage of relevant high frequency information in the image (similar to Cutout). Finally, we show that Patch Gaussian can be used in conjunction with other regularization methods and data augmentation policies such as AutoAugment, and improves performance on the COCO object detection benchmark.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Safety, Robustness & Interpretability

Citations in the corpus

On Robustness and Transferability of Convolutional Neural Networks (2020, background): “Using better data augmentation can improve performance on data with synthetic noise [29, 43].”
AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty (2019, background): “Patch Gaussian augments data with Gaussian noise applied to a randomly chosen portion of an image (Lopes et al., 2019).”
Adversarial Examples Improve Image Recognition (2019, background): “, applying masking out [5] or adding Gaussian noise [22] to regions in images, or mixing up pairs of images and their labels in a convex manner [49].”
Learning Data Augmentation Strategies for Object Detection (2019, methodology): “Instead of cropping to focus on parts of the image, some methods randomly erase or add noise to patches of images for improved accuracy [9, 53, 13], robustness [50, 12], or both [29].”
Quality-Agnostic Image Recognition via Invertible Decoder (2021, background): “Another research line related to data augmentation attempts to inject noise to input images (or patches) for improving model robustness to noise corruptions as well as other types of corruptions [33, 38].”
A Simple Way to Make Neural Networks Robust Against Diverse Image Corruptions (2020, methodology): “We evaluate all proposed methods for ImageNet-C on the ResNet50 architecture for better comparability to previous methods, e.g. [9], [19], [45].”
A Group-Theoretic Framework for Data Augmentation (2019, background): “For example, a line of work proposes to add random or adversarial noise when training [11, 51, 26, 74, 8, 49, 73, 37] and argue that this can lead to a form of regularization [63, 34, 35, 18, 71].”
The Origins and Prevalence of Texture Bias in Convolutional Neural Networks (2019, background): “Recent work has shown that data augmentation can also improve out-of-distribution robustness [33, 102, 65, 45, 78].”
Causally motivated shortcut removal using auxiliary labels (2021, methodology): “Discouraging shortcut learning using data augmentation (e.g., rotation, translation, cropping) has been suggested by multiple authors [Hendrycks et al., 2020, Yin et al., 2019, Lyle et al., 2020, Lopes et al., 2019, Cubuk et al., 2018].”
RandAugment: Practical data augmentation with no separate search (2019, methodology): “Some methods randomly erase or add noise to patches of images for increased validation accuracy [8, 54], robustness [45, 51, 11], or both [31].”
Adversarial Robustness with Semi-Infinite Constrained Learning (2021, background): “For instance, for λ = N ( 0 , Σ ) , (PII) becomes Indeed, the authors of [84, 85] suggest that Gaussian data augmentation can signiﬁcantly improve generalization, particularly the augmentations are applied using patches [84].”
Increasing the robustness of DNNs against image corruptions by playing the Game of Noise (2020, methodology): “We evaluate all proposed methods for ImageNet-C on the ResNet50 architecture for better comparability to previous methods, e.g. [Geirhos et al., 2019, Lopes et al., 2019, Zhang, 2019].”
Affinity and Diversity: Quantifying Mechanisms of Data Augmentation (2020, methodology): “Furthermore, methods do not always transfer across datasets—Cutout, for example, is useful on CIFAR-10 and not on ImageNet [5].”