GSoC Project: Implement the command translation layer inside of PySyft which will allow for PySyft Tensorflow plans to be converted to PyTorch

[cereallarceny]

Problem: Essentially we’ll already have a translation function that exists for Javascript - this project simply brings the same logic to Python. This may not end up serving a permanent purpose in PySyft, but we think that this would allow for a developer to create a plan using PySyft for TensorFlow and still have training available for mobile phones. This would be an incredible achievement since currently we’re scoped to only allow the training of PyTorch models on mobile devices (since they require the use of TorchScript). If a developer wanted to write their PySyft model using TensorFlow, they would then need to convert this plan to use PyTorch, before ultimately hosting the plan on PyGrid as a TorchScript.

Solution: We are building a command translation library for the syft.js web worker library called "Threepio". This project will need to be extended to support TensorFlow (Python) bindings. By this point, it should already have bindings for PyTorch and TensorFlow.js.

Requirements:

• Willing to work with the Web & Mobile team and be mentored by @Nolski, the author of Threepio
• Familiarity with multiple deep learning frameworks: PyTorch, TensorFlow, and TensorFlow.js (NumPy is a bonus!)
• Good knowledge and understanding of the problems and challenges described in the Federated Learning roadmap (questions can be directed to @cereallarceny)

Difficulty: Medium

It’s worth noting that this project is considered optional “extra credit” for the Federated Learning roadmap: https://github.com/OpenMined/Roadmap/blob/master/web_and_mobile_team/projects/federated_learning.md

[github-actions[bot]]

This issue has been marked stale because it has been open 30 days with no activity. Leave a comment or remove the stale label to unmark it. Otherwise, this will be closed in 7 days.