English

A General Approach to Adding Differential Privacy to Iterative Training Procedures

Machine Learning 2019-03-05 v2 Machine Learning

Abstract

In this work we address the practical challenges of training machine learning models on privacy-sensitive datasets by introducing a modular approach that minimizes changes to training algorithms, provides a variety of configuration strategies for the privacy mechanism, and then isolates and simplifies the critical logic that computes the final privacy guarantees. A key challenge is that training algorithms often require estimating many different quantities (vectors) from the same set of examples --- for example, gradients of different layers in a deep learning architecture, as well as metrics and batch normalization parameters. Each of these may have different properties like dimensionality, magnitude, and tolerance to noise. By extending previous work on the Moments Accountant for the subsampled Gaussian mechanism, we can provide privacy for such heterogeneous sets of vectors, while also structuring the approach to minimize software engineering challenges.

Keywords

Cite

@article{arxiv.1812.06210,
  title  = {A General Approach to Adding Differential Privacy to Iterative Training Procedures},
  author = {H. Brendan McMahan and Galen Andrew and Ulfar Erlingsson and Steve Chien and Ilya Mironov and Nicolas Papernot and Peter Kairouz},
  journal= {arXiv preprint arXiv:1812.06210},
  year   = {2019}
}

Comments

Presented at NeurIPS 2018 workshop on Privacy Preserving Machine Learning; Companion paper to TensorFlow Privacy OSS Library

R2 v1 2026-06-23T06:43:14.479Z