BriefGPT.xyz
Dec, 2018
迭代训练程序添加差分隐私的一般方法
A General Approach to Adding Differential Privacy to Iterative Training Procedures
HTML
PDF
H. Brendan McMahan, Galen Andrew
TL;DR
本文介绍了一种模块化方法，可以最小化对训练算法的更改，提供各种隐私机制的配置策略，并隔离和简化计算最终隐私保证的关键逻辑，以解决在隐私敏感数据集上训练机器学习模型所面临的实际挑战。
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
→