My attempt to implement Decoupled Neural Interfaces using Synthetic Gradients
The models are built using Tensorflow 1.0.0
A working decoupled model using synthetic gradients tested on a fully-connected neural net trained on MNIST can be found in:
tensorflow/scripts_dev/experiments/exp06.py
The original non-decoupled network trained on MNIST can be found in:
tensorflow/scripts_dev/working_mnist.ipynb