Agnostic Equilibrium Propagation

This repository contains code to produce simulations of Agnostic Equilibrium Propagation on Hopfield-like networks.

The code uses PyTorch, TorchVision and TensorBoard.

Getting started

• Download the code from GitHub:

git clone https://github.com/bscellier/agnostic-equilibrium-propagation
cd agnostic-equilibrium-propagation

• To train a Hopfield-like network (with 1 hidden layer by default) on MNIST with Agnostic Eqprop, run the command:

python run.py --dataset='MNIST' --architecture='1h' --method='centered' --verbose

• To run Tensorboard, use the command:

tensorboard --logdir=runs/MNIST/1h/

• Once the model is trained, start the GUI by running the command:

python run_gui.py --dataset='MNIST' --architecture='1h'