I tried to implement Deepminds 2017 paper on synthetic gradients and I am trying to combine synthetic gradients to computer vision tasks to improve training speed as well as accuracy. The great visulaization and code was all made possible because of Trask, the explanation has been modified a bit and I also took some inspiration from Siraj's video, so do give it a like, I wrote the backprop code and the code to implement the synthetic gradients using some parts from trask's repo. as they were written far cleanely than mine. So for it, please visit his site and github repo and give an upvote Link:https://iamtrask.github.io/2017/03/21/synthetic-gradients/.