Morten Dahl

haven't been able to find an optimized protocol for this; the log round standard protocol of course works (even for Pond) but worried about the graph size we get from...

Note to self: find a way where the federated protocol simply wraps an existing model on the model owner?

Design and discussion has moved into a RFC: https://github.com/tf-encrypted/rfcs/pull/1

@yanndupis @jvmancuso the more I look into this the more I think we should simply stick with the FL example we have, but upgrade it to use TF Keras and...

> For example [here], we would like to avoid defining manually the model weights. Yes, fully agree. > It would be nice to kick the training process with `model.fit(data_sources, epochs=10)`...

Could you provide a few more details @jvmancuso?

Any ideas @justin1121 @yanndupis? I can pick it up tomorrow morning.

> I'm confused why anything is running on the 5th CPU player though especially see as that CPU doesn't even exist. not sure that's what you're asking, but `LocalConfig` simulates...

> I'd expect the available devices to to show all devices available not just ones configured with LocalConfig. I believe the problem is that the session has *not* been configured...

Related: would like to essentially revert https://github.com/tf-encrypted/tf-encrypted/pull/512/commits/efdf6049c4f1bcd81ef79edbbbb83aa2c208b6d6