Privacy Preserving Federated Ranker

Proof of concept privacy preserving collaborative filtering recommendation system. Building on Tensorflow Federated, the server and client independently train their own Matrix Factorization models, using only a subset of shared latent factors. The server aggregates client model updates and averages them into the global model that future clients will use.

Getting Started

1. Generate a self-signed certificate & key:

cd data
openssl req -x509 -newkey rsa:4096 -keyout key.pem -out cert.pem -days 365

2. Run the server & client via:

python src/server.py
python src/client.py

3. To run tests

python -m unittest discover tests