DRO-Augment：用 Wasserstein 分布鲁棒优化改善 Mixup 校准

View PDF HTML (experimental)

Abstract:In many real-world applications, ensuring the robustness and stability of deep neural networks (DNNs) is crucial, particularly for image classification tasks that encounter various input perturbations. While Mixup-based data augmentation techniques have been widely adopted to enhance the resilience of trained models against such perturbations, our experiments reveal an important corruption robustness-calibration trade-off: stronger Mixup-based augmentation can improve robustness against corrupted data while substantially increasing expected calibration error (ECE). To address this challenge, we introduce DRO-Augment, a framework that integrates Wasserstein Distributionally Robust Optimization (W-DRO) with various Mixup-based data augmentation strategies to mitigate this trade-off. Our method substantially reduces ECE under strong Mixup-based augmentation while largely preserving corruption accuracy across CIFAR-10, CIFAR-100, CIFAR-10-C, and CIFAR-100-C. On the theoretical side, we establish novel generalization error bounds for neural networks trained using a variation-regularized loss function with augmented data, closely related to the W-DRO problem. Furthermore, we introduce a refined CIFAR-C benchmark that corrects inconsistencies in corruption intensities, providing a more reliable evaluation for future robustness research.

Comments:	21 pages
Subjects:	Machine Learning (stat.ML); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
Cite as:	arXiv:2506.17874 [stat.ML]
	(or arXiv:2506.17874v3 [stat.ML] for this version)
	https://doi.org/10.48550/arXiv.2506.17874 arXiv-issued DOI via DataCite

Submission history

From: Jiaming Hu [view email]
[v1] Sun, 22 Jun 2025 02:18:03 UTC (741 KB)
[v2] Tue, 24 Jun 2025 21:04:53 UTC (734 KB)
[v3] Tue, 6 Oct 2026 03:25:38 UTC (42 KB)