Skip to content

Implementation of Hinton's forward-forward (FF) algorithm - an alternative to back-propagation

Notifications You must be signed in to change notification settings

haarisr/pytorch_forward_forward

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 

Repository files navigation

pytorch_forward_forward

Implementation of forward-forward (FF) training algorithm - an alternative to back-propagation

Below is my understanding of the FF algorithm presented at Geoffrey Hinton's talk at NeurIPS 2022.
The conventional backprop computes the gradients by successive applications of the chain rule, from the objective function to the parameters. FF, however, computes the gradients locally with a local objective function, so there is no need to backpropagate the errors.

The local objective function is designed to push a layer's output to values larger than a threshold for positive samples and to values smaller than a threshold for negative samples.

A positive sample $s$ is a real datapoint with a large $P(s)$ under the training distribution.
A negative sample $s'$ is a fake datapoint with a small $P(s')$ under the training distribution.

Among the many ways of generating the positive/negative samples, for MNIST, we have:
Positive sample $s = merge(x, y)$, the image and its label
Negative sample $s' = merge(x, y_{random})$, the image and a random label

After training all the layers, to make a prediction for a test image $x$, we find the pair $s = (x, y)$ for all $0 \leq y < 10$ that maximizes the network's overall activation.

With this implementation, the training and test accuracy on MNIST are:

> python main.py
train error: 0.06520003080368042
test error: 0.09890002012252808

About

Implementation of Hinton's forward-forward (FF) algorithm - an alternative to back-propagation

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%