Backpropagation about rwa HOT 3 CLOSED

It is in the old version of the manuscript: https://arxiv.org/abs/1703.01253v2

I removed it from the updated manuscript because I want to double check the equations. The idea may be flawed. If it works, I will put the equations back in. TensorFlow will use normal backpropagation to the fit the model without needing the recursive gradient equations

from rwa.

Ok Thanks.That solves the issue.

from rwa.

Just to follow up on the issue of computing the gradients by a recursive forward pass. I did not have the equations written correctly in that earlier version of the manuscript but I think the concept will work. I am playing around with the idea some more, and hope to publish a follow up manuscript if the idea works,

from rwa.