There are 0 repository under synthetic-gradients topic.
Implement Decoupled Neural Interfaces using Synthetic Gradients in Pytorch
Decoupled Neural Interfaces Using Synthetic Gradients - under develeopment
TensorFlow implementation of "Decoupled Neural Interfaces using Synthetic Gradients"
Predicting gradient (of perceptual loss) without doing a backward pass
Decoupled Neural Interfaces (Jaderberg et al. 2017) mini-package for easy integration with pytorch RNNs
Tests convergence of various estimators. In particular, PD and SF/REINFORCE are tested for gaussian and bernoulli latent variables and Synthetic gradients are tested in MNIST Classification (DNI).
This repo contains a simple neural network to predict outcome of a clash royale game, this project is done solely to compare the execution speed vs accuracy of the new training method (synthetic gradients) vs the ubiquotous backpropogation algorithm