Support to FastAI

[rondey]

Feature Description

It could be great write private neural networks using the modern best practices implemented in FastAI. I tried PySyft 0.2 alongside with FastAI (v1) and they are not perfectly compatible each other. Do you think is possible make compatible the new PySyft 0.3 with FastAI v2?

[madhavajay]

It totally could be possible. I know a few people from the FastAI Audio group who might be interested in helping. @rbracco ?

[madhavajay]

@rondey I would also highly recommend posting this on the FastAI Discord / Forums as well and see if @jph00 is potentially interested in giving it some official sanctioning.

[rondey]

@madhavajay Thank you so much, I have created a new post on the FastAI forum.

[jph00]

Looking forward to seeing how you get along with this! :D

[madhavajay]

@jph00 I'll consider that officially sanctioned 👍🏼

@rondey I am a big FastAI fan myself but we don't have the bandwidth to start this right now while we are building support for all the other libraries needed to provide the PPML functionality you mentioned in the FastAI forum post. 😂. However if you are interested in opening an initial PR on this I can walk you through the process. ❤️

[rondey]

I like the idea, but actually I have a paper to complete (I'm using PySyft for the implementation).

[madhavajay]

@rondey No worries, I am sure we will get to very soon, but we just have a few priorities of our own as well. 😊 Good luck with the paper and thanks for using PySyft in the implementation. ❤️

[rondey]

Thank you! I perfectly understand your priorities

[madhavajay]

This is going into a backlog, don't worry, we will be doing this. 😊