Improving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation

Abstract

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.

References in corpus (19)

• mixup: Beyond Empirical Risk Minimization
• Improved Regularization of Convolutional Neural Networks with Cutout
• Wide Residual Networks
• Random Erasing Data Augmentation
• AutoAugment: Learning Augmentation Policies from Data
• DropBlock: A regularization method for convolutional networks
• Do ImageNet Classifiers Generalize to ImageNet?
• Adversarial Examples Are Not Bugs, They Are Features
• Shake-Shake regularization
• Why do deep convolutional networks generalize so poorly to small image transformations?
• Motivating the Rules of the Game for Adversarial Example Research
• Adversarial Examples Are a Natural Consequence of Test Error in Noise
• Do CIFAR-10 Classifiers Generalize to CIFAR-10?
• Benchmarking Neural Network Robustness to Common Corruptions and Surface Variations
• Learning Robust Representations by Projecting Superficial Statistics Out
• Analyzing noise in autoencoders and deep networks
• Excessive Invariance Causes Adversarial Vulnerability
• Using Videos to Evaluate Image Model Robustness
• Intriguing Properties of Adversarial Examples

Cited by in corpus (41)

• AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
• Self-training with Noisy Student improves ImageNet classification
• The Origins and Prevalence of Texture Bias in Convolutional Neural Networks
• Evaluating Prediction-Time Batch Normalization for Robustness under Covariate Shift
• An overview of mixing augmentation methods and augmentation strategies
• Affinity and Diversity: Quantifying Mechanisms of Data Augmentation
• Inspect, Understand, Overcome: A Survey of Practical Methods for AI Safety
• Adversarial Examples Improve Image Recognition
• Test-Time Adaptation to Distribution Shift by Confidence Maximization and Input Transformation
• Efficient Certified Defenses Against Patch Attacks on Image Classifiers
• Robustness properties of Facebook's ResNeXt WSL models
• Viewmaker Networks: Learning Views for Unsupervised Representation Learning
• Boundary thickness and robustness in learning models
• Does enhanced shape bias improve neural network robustness to common corruptions?
• Robust Image Classification Using A Low-Pass Activation Function and DCT Augmentation
• On Interaction Between Augmentations and Corruptions in Natural Corruption Robustness
• SCARF: Self-Supervised Contrastive Learning using Random Feature Corruption
• PDA: Progressive Data Augmentation for General Robustness of Deep Neural Networks
• Bit Error Robustness for Energy-Efficient DNN Accelerators
• DeeperForensics Challenge 2020 on Real-World Face Forgery Detection: Methods and Results
• Evaluating COPY-BLEND Augmentation for Low Level Vision Tasks
• Causally motivated Shortcut Removal Using Auxiliary Labels
• Rethinking Recurrent Neural Networks and Other Improvements for Image Classification
• Meta Approach to Data Augmentation Optimization
• Discovering conservation laws from trajectories via machine learning
• Towards Corruption-Agnostic Robust Domain Adaptation
• Anti-aliasing Deep Image Classifiers using Novel Depth Adaptive Blurring and Activation Function
• Pyramid Adversarial Training Improves ViT Performance
• Does Data Augmentation Benefit from Split BatchNorms
• Learning Visual Representations for Transfer Learning by Suppressing Texture
• Meta Adversarial Training against Universal Patches
• Defending Against Image Corruptions Through Adversarial Augmentations
• Random Shadows and Highlights: A new data augmentation method for extreme lighting conditions
• Cooperative Training and Latent Space Data Augmentation for Robust Medical Image Segmentation
• RoMA: Robust Model Adaptation for Offline Model-based Optimization
• Learning Augmentation Distributions using Transformed Risk Minimization
• Exploring Frequency Domain Interpretation of Convolutional Neural Networks
• Untapped Potential of Data Augmentation: A Domain Generalization Viewpoint
• Combining Different V1 Brain Model Variants to Improve Robustness to Image Corruptions in CNNs
• Robust Face Verification via Disentangled Representations
• Balancing Robustness and Sensitivity using Feature Contrastive Learning