IPMix: Label-Preserving Data Augmentation Method for Training Robust Classifiers
This work addresses the need for robust deep neural networks in real-world scenarios where data distributions shift, offering a method that enhances robustness without compromising accuracy, though it is incremental as it builds on existing data augmentation techniques.
The paper tackles the problem of training robust classifiers without sacrificing clean accuracy by proposing IPMix, a label-preserving data augmentation method that integrates image-, patch-, and pixel-level augmentations; it outperforms state-of-the-art methods on corruption robustness benchmarks like CIFAR-C and ImageNet-C and improves other safety measures such as adversarial robustness and anomaly detection.
Data augmentation has been proven effective for training high-accuracy convolutional neural network classifiers by preventing overfitting. However, building deep neural networks in real-world scenarios requires not only high accuracy on clean data but also robustness when data distributions shift. While prior methods have proposed that there is a trade-off between accuracy and robustness, we propose IPMix, a simple data augmentation approach to improve robustness without hurting clean accuracy. IPMix integrates three levels of data augmentation (image-level, patch-level, and pixel-level) into a coherent and label-preserving technique to increase the diversity of training data with limited computational overhead. To further improve the robustness, IPMix introduces structural complexity at different levels to generate more diverse images and adopts the random mixing method for multi-scale information fusion. Experiments demonstrate that IPMix outperforms state-of-the-art corruption robustness on CIFAR-C and ImageNet-C. In addition, we show that IPMix also significantly improves the other safety measures, including robustness to adversarial perturbations, calibration, prediction consistency, and anomaly detection, achieving state-of-the-art or comparable results on several benchmarks, including ImageNet-R, ImageNet-A, and ImageNet-O.