This repo contains various machine learning tools based on the secure multi-party framework FRESCO. These allows multiple parties to train and/or evaluate machine learning models on their combined data without sharing their data with the other parties.
The tools currently include:
- Neural networks
- Decision trees
- Federated learning
- Support vector machines
- Logistic regression
The easist way to see how each tool is used is to see the corresponding unit tests.
The repo contains a demo of federated learning on the MNIST handwritten digits dataset where all parties run locally for 10 global epochs. Build it with mvn packageand run ./launchscript.sh [number of parties] [number of local examples]. Example parameters could be 3 parties and 5.000 local examples each.