tags	dataset	framework
dp security fds	CIFAR-10	opacus torch

Training with Sample-Level Differential Privacy using Opacus Privacy Engine

In this example, we demonstrate how to train a model with differential privacy (DP) using Flower. We employ PyTorch and integrate the Opacus Privacy Engine to achieve sample-level differential privacy. This setup ensures robust privacy guarantees during the client training phase. The code is adapted from the PyTorch Quickstart example.

For more information about DP in Flower please refer to the tutorial. For additional information about Opacus, visit the official website.

Environments Setup

Start by cloning the example. We prepared a single-line command that you can copy into your shell which will checkout the example for you:

git clone --depth=1 https://github.com/adap/flower.git && mv flower/examples/opacus . && rm -rf flower && cd opacus

This will create a new directory called opacus containing the following files:

-- pyproject.toml
-- client.py
-- server.py
-- README.md

Installing dependencies

Project dependencies are defined in pyproject.toml. Install them with:

pip install .

Run Flower with Opacus and Pytorch

1. Start the long-running Flower server (SuperLink)

flower-superlink --insecure

2. Start the long-running Flower clients (SuperNodes)

Start 2 Flower SuperNodes in 2 separate terminal windows, using:

flower-client-app client:appA --insecure

flower-client-app client:appB --insecure

Opacus hyperparameters can be passed for each client in ClientApp instantiation (in client.py). In this example, noise_multiplier=1.5 and noise_multiplier=1 are used for the first and second client respectively.

3. Run the Flower App

With both the long-running server (SuperLink) and two clients (SuperNode) up and running, we can now run the actual Flower App:

flower-server-app server:app --insecure