pytorch-dp: Train PyTorch models with Differential Privacy

Main build:

Release build: (v0.1-beta.1)

pytorch-dp is a library that enables training PyTorch models with differential privacy. It supports training with minimal code changes required on the client, has little impact on training performance and allows the client to online track the privacy budget expended at any given moment.

PyTorch-DP is currently a preview beta and under active development!

Target audience

This code release is aimed at two target audiences:

ML practicioners will find this code a gentle introduction to training a model with differential privacy as it requires minimal code changes.
Differential Privacy scientists will find this code easy to experiment and tinker with, allowing them to focus on what matters.

Installation

pip:

pip install pytorch-dp

From source:

git clone https://github.com/facebookresearch/pytorch-dp.git
cd pytorch-dp
pip install -e .

Getting started

To train your model with differential privacy, all you need to do is to declare a PrivacyEngine and attach it to your optimizer before running, eg:

model = Net()
optimizer = SGD(model.parameters(), lr=0.05)
privacy_engine = PrivacyEngine(
    model,
    batch_size,
    sample_size,
    alphas=[1, 10, 100],
    noise_multiplier=1.3,
    max_grad_norm=1.0,
)
privacy_engine.attach(optimizer)
# Now it's business as usual

The MNIST example contains an end to end run.

Contributing

See the CONTRIBUTING file for how to help out.

References

License

This code is released under Apache 2.0, as found in the LICENSE file.