Opacus: User-Friendly Differential Privacy Library in PyTorch

TL;DR

Opacus is an easy-to-use, efficient PyTorch library that enables practitioners to implement differential privacy in deep learning models with minimal code changes, supporting various layers and improving gradient computation efficiency.

Contribution

The paper introduces Opacus, a user-friendly, flexible, and high-performance differential privacy library for PyTorch that simplifies privacy-preserving model training.

Findings

Supports a wide range of neural network layers out of the box.

Provides higher efficiency through batched per-sample gradient computation.

Benchmarks show competitive performance with existing frameworks.

Abstract

We introduce Opacus, a free, open-source PyTorch library for training deep learning models with differential privacy (hosted at opacus.ai). Opacus is designed for simplicity, flexibility, and speed. It provides a simple and user-friendly API, and enables machine learning practitioners to make a training pipeline private by adding as little as two lines to their code. It supports a wide variety of layers, including multi-head attention, convolution, LSTM, GRU (and generic RNN), and embedding, right out of the box and provides the means for supporting other user-defined layers. Opacus computes batched per-sample gradients, providing higher efficiency compared to the traditional "micro batch" approach. In this paper we present Opacus, detail the principles that drove its implementation and unique features, and benchmark it against other frameworks for training models with differential privacy as well as standard PyTorch.