New framework DRO-Augment improves neural network calibration with Mixup augmentation

Researchers have developed DRO-Augment, a new framework designed to improve the calibration of deep neural networks when using Mixup-based data augmentation. This method integrates Wasserstein Distributionally Robust Optimization (W-DRO) to address the trade-off where stronger augmentation improves robustness against corrupted data but increases calibration error. DRO-Augment significantly reduces calibration error while maintaining accuracy on datasets like CIFAR-10 and CIFAR-100, and also introduces a refined CIFAR-C benchmark for future research. AI

IMPACT This research could lead to more reliable and accurately calibrated deep learning models, particularly in image classification tasks facing data corruption.

RANK_REASON The cluster contains an academic paper detailing a new optimization framework for deep neural networks. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• CIFAR-10
• CIFAR-100
• CIFAR-100-C
• CIFAR-10-C
• DRO-Augment
• Jiaming Hu
• MIXUP
• Wasserstein distributionally robust optimization

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →