sajjadmanal/syft-js-worker

Javascript worker for PySyft

Syft Web Worker

This project is created for Udacity's Secure and Private AI challenge course. It explores possibility to have PySyft worker running in the web browser, which potentially expands PySyft functionality to a very large user-base. The back-end for tensor operations in browser is tf.js.

The current project's state is 'proof of concept' where tensors can be moved to/from browser and a limited set of tensor operations is possible. Further direction is to try to use web client for federated learning.

Demo

Follow this jupyter notebook, it works in Colab!

For local installation:

pip install -Ur requirements.txt
python run_socketio_server.py

Update index.html with server's URL (e.g. http://localhost:5000).

Work in progress

There's zero infrastructure work done (e.g. dependency management, building, etc.), and javascript is tested in Chrome only.

Acknowledgments

• syft.js - inspiration :)
• PySyft Android Worker - server script and some PySyft protocol details
• Numpy to array code borrowed from: https://gist.github.com/nvictus/88b3b5bfe587d32ac1ab519fd0009607
• Array to numpy code borrowed from: https://github.com/propelml/tfjs-npy/blob/master/npy.ts
• Thanks Udacity's Secure and Private AI challenge course and Andrew Trask for great content