DP-Mix: Mixup-based Data Augmentation for Differentially Private Learning

TL;DR

This paper introduces DP-Mix, two novel data augmentation methods tailored for differentially private learning, significantly improving model performance by combining mixup with self-augmentation and synthetic data from diffusion models.

Contribution

The paper proposes two new mixup-based data augmentation techniques specifically designed for differentially private learning, addressing the incompatibility of traditional augmentation methods.

Findings

DP-Mix_Self achieves state-of-the-art results across datasets.

DP-Mix_Diff further enhances performance using synthetic data.

Both methods outperform existing private learning approaches.

Abstract

Data augmentation techniques, such as simple image transformations and combinations, are highly effective at improving the generalization of computer vision models, especially when training data is limited. However, such techniques are fundamentally incompatible with differentially private learning approaches, due to the latter's built-in assumption that each training image's contribution to the learned model is bounded. In this paper, we investigate why naive applications of multi-sample data augmentation techniques, such as mixup, fail to achieve good performance and propose two novel data augmentation techniques specifically designed for the constraints of differentially private learning. Our first technique, DP-Mix_Self, achieves SoTA classification performance across a range of datasets and settings by performing mixup on self-augmented data. Our second technique, DP-Mix_Diff, further improves performance by incorporating synthetic data from a pre-trained diffusion model into the mixup process. We open-source the code at https://github.com/wenxuan-Bao/DP-Mix.