Implementation of Relax method for Bernoulli random variable
Implementation of relax method for Bernoulli random variable from the paper Backpropagation through the Void: Optimizing control variates for black-box gradient estimation, Will Grathwohl, Dami Choi, Yuhuai Wu, Geoffrey Roeder, David Duvenaud ,https://arxiv.org/abs/1711.00123
The implementation is a copy-cat from https://github.com/duvenaud/relax
I have tried to improve handling coefficients of \theta and \phi.
This could use some readme/walkthrough for someone who isn't familiar with the area. Are there still open Flux API questions? I can try to help with those things if they are narrowed down a bit.
Hi Mike,
I would like to get back to this again. I can write a readme explaining, what the method does. The only mild issue is that I do not know, how to include latex into readme. It would be the best, if I can add some equations.
I suggest avoiding latex in the readme but you can add some in the code using jupyter's syntax. That will work if the script is converted to a notebook via literate.jl. The readme can just be an overview with the script giving a more detailed walkthrough.