Research map / Federated learning and privacy / Differential privacy
DP-Mix: Mixup-based Data Augmentation for Differentially Private Learning
Wenxuan Bao, Francesco Pittaluga, Vijay Kumar b g, Vincent Bindschaedler
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply mixup via microbatch size 2 on top of DP-SGD (ablation) | DP-SGD · CIFAR-10 WRN-16-4 eps=8 | test accuracy: 72.5 -> 49.7 | Hurt |
| add mixup within microbatch of 2 to DP-SGD (ablation) | DP-SGD · CIFAR-10 microbatch 2 | test accuracy: 49.7 -> 50.1 | No clear effect |
| replace Self-Aug with Self-Mix | Self-Aug · CIFAR-10 from scratch eps=8 | test accuracy: 78.7 -> 79.8 | Improved |
| replace Self-Aug with Self-Mix | Self-Aug · CIFAR-100 from scratch eps=8 | test accuracy: 39.2 -> 40.6 | Improved |
| replace Self-Aug with Self-Mix | Self-Aug · CIFAR-10 from scratch eps=1 | test accuracy: 56.8 -> 57.2 | Improved |
| replace Self-Aug with Self-Mix | Self-Aug · ViT-B-16 CIFAR-100 eps=1 | test accuracy: 79.3 -> 81.8 | Improved |
| add diffusion samples via Diff-Mix instead of Self-Mix | Self-Mix · ViT-B-16 Caltech256 eps=1 | test accuracy: 81.2 -> 89.7 | Improved |
| add diffusion samples via Diff-Mix instead of Self-Mix | Self-Mix · ViT-B-16 Oxford Pet eps=1 | test accuracy: 72.5 -> 83.2 | Improved |
| add diffusion samples via Diff-Mix instead of Self-Mix | Self-Mix · ViT-B-16 EuroSAT eps=1 | test accuracy: 94.3 -> 92.6 | Hurt |
| add diffusion samples via Diff-Mix instead of Self-Mix | Self-Mix · ViT-B-16 CIFAR-10 eps=8 | test accuracy: 97.6 -> 97.3 | Hurt |
| add Sobel filter augmentation to Self-Aug (ablation) | Self-Aug · ViT-B-16 CIFAR-100 eps=1 | test accuracy: 79.3 -> 9.5 | Hurt |
| add Gaussian blur augmentation to Self-Aug (ablation) | Self-Aug · ViT-B-16 CIFAR-100 eps=1 | test accuracy: 79.3 -> 80.0 | No clear effect |
Closest papers by meaning
- Differentially Private Fine-Tuning of Diffusion Models (ICCV 2025 · similarity 0.87)
- Differentially Private Prototypes for Imbalanced Transfer Learning (AAAI 2025 · similarity 0.86)
- Catch-Up Mix: Catch-Up Class for Struggling Filters in CNN (AAAI 2024 · similarity 0.86)
- In-distribution Public Data Synthesis with Diffusion Models for Differentially Private Image Classification (CVPR 2024 · similarity 0.86)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.86)
- Enhance Image Classification via Inter-Class Image Mixup with Diffusion Model (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
