A General Approach to Adding Differential Privacy to Iterative Training Procedures

2018/12/15 by H. Brendan McMahan, McMahan, H. Brendan, Galen Andrew +12 · 7 citations
Computer Science · Mathematics · #Advanced Neural Network Applications #Adversarial Robustness in Machine Learning #Privacy-Preserving Technologies in Data #cs.LG #stat.ML

paper · pdf · doi:10.48550/arxiv.1812.06210

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

arxiv created 2019/03/04 · arxiv updated 2019/03/05

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.

Cited by

• Scaling Trust in Quantum Federated Learning: A Multi-Protocol Privacy Design
• On Optimal Hyperparameters for Differentially Private Deep Transfer Learning
• Parameter-Efficient and Personalized Federated Training of Generative Models at the Edge
• FedRP: A Communication-Efficient Approach for Differentially Private Federated Learning Using Random Projection
• Rethinking Layer-wise Gaussian Noise Injection: Bridging Implicit Objectives and Privacy Budget Allocation
• Differentially Private Secure Multiplication with Erasures and Adversaries
• Releasing Differentially Private Event Logs Using Generative Models

Related

• Three Tools for Practical Differential Privacy
• Too Good to be True? Turn Any Model Differentially Private With DP-Weights
• Deep Learning with Differential Privacy
• Data Heterogeneity Differential Privacy: From Theory to Algorithm
• Privacy Auditing with One (1) Training Run