Skip to content

Latest commit

 

History

History
34 lines (25 loc) · 1.12 KB

README.md

File metadata and controls

34 lines (25 loc) · 1.12 KB
Raw

Mesh Neural Networks

This repository contains a Tensorflow 2.1 implementation of Mesh Neural Network, based on article Formal derivation of Mesh Neural Networks with their Forward-Only gradient Propagation.

Overview

Unlike feed-forward neural network, grouping neurons by layers, where only neighboring layers are connected, mesh neural networks represents network as a weighted graph, where all neurons (except input ones) are connected densely.

mesh_nn_as_graph_from article

How to run

This repository contain two train scripts on MNIST: one for mesh neural network

python3 train_mesh_nn.py

and one for feed-forward neural network

python3 train_dense.py

Both scripts are constructed to have the same hyperparameters and more or less the same networks complexity. Training process could be watched via tensorboard:

tensorboard --logdir logs

tensorboard_screenshot

Requirements

  • tensorflow == 2.1
  • numpy >= 1.12.0