tf-encrypted Depthwise separable 2D convolution

Hi,

Is there a plan to add support for depthwise separable 2D convolution (DepthwiseConv2D) ?

Thanks

Apr 30 '19 11:04 avitalsh

Yes, it's on our roadmap, although we've prioritized a few other items (in particular, more robust support for Keras layers and models). @yanndupis and I have some other things on our roadmap ahead of that, but we had hoped to get to it in the next couple weeks.

If you'd like to submit a PR adding it to the Pond protocol, we'd be more than happy to work with you to merge it!

Apr 30 '19 18:04 jvmncs

@jvmancuso is this something we should bump up and look at within a week or two?

May 07 '19 08:05 mortendahl

This would be great to work on after the keras work is done. The forward pass is easy, but the backward pass can be quite tricky.

May 07 '19 12:05 jvmncs

@jvmancuso Hey, is this planned to be added sometime soon?

Jun 25 '19 08:06 avitalsh

We've had plans to do so, but it's been a bit neglected -- would you find it useful to have it?

If you'd like to add it, we'd be more than happy to help you with the implementation and get it merged.

Jun 25 '19 14:06 jvmncs