Differentially Private Mixing of Public Datasets Improves Private Learning

arXiv:2610.06636 · cs.LG, cs.AI, cs.CR · Submitted 2026-10-05 · Read on arXiv

cs.LG, cs.AI, cs.CR

Submitted: 2026-10-05

Updated: 2026-10-05

Project page: https://stanfordmlgroup.github.io/competitions/chexpert

Terminology

• Differentially Private — 1× in this paper · explained in 3 episodes

Sources

• Mind the Privacy Unit! User-Level Differential Privacy for Language Model Fine-Tuning
• Unlocking High-Accuracy Differentially Private Image Classification through Scale
• Training on the Test Task Confounds Evaluation and Emergence
• Why Is Public Pretraining Necessary for Private Model Training?
• The Pile: An 800GB Dataset of Diverse Text for Language Modeling
• Choosing Public Datasets for Private Machine Learning via Gradient Subspace Distance
• CheXpert: A Large Chest Radiograph Dataset with Uncertainty Labels and Expert Comparison
• Auditing Differentially Private Machine Learning: How Private is Private SGD?
• MIMIC-CXR-JPG, a large publicly available database of labeled chest radiographs
• The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text
• RegMix: Data Mixture as Regression for Language Model Pre-training
• Large Scale Transfer Learning for Differentially Private Image Classification
• R'enyi Differential Privacy of the Sampled Gaussian Mechanism
• OpenWebMath: An Open Dataset of High-Quality Mathematical Web Text
• How to DP-fy Your Data: A Practical Guide to Generating Synthetic Data With Differential Privacy
• CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning
• MixMin: Finding Data Mixtures via Convex Minimization
• Position: Considerations for Differentially Private Learning with Large-Scale Public Pretraining
• DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining
• Opacus: User-Friendly Differential Privacy Library in PyTorch

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks