The Forward-Forward Algorithm: Some Preliminary Investigations

TL;DR

This paper introduces the Forward-Forward algorithm, a new neural network training method that replaces backpropagation with two forward passes using positive and negative data, showing promising initial results.

Contribution

The paper proposes the Forward-Forward algorithm, a novel learning procedure that simplifies neural network training by eliminating backward passes and enabling offline negative data generation.

Findings

Works well on small problems, warranting further research.

Allows offline negative data generation, simplifying training.

Potential for pipelined processing without storing activities.

Abstract

The aim of this paper is to introduce a new learning procedure for neural networks and to demonstrate that it works well enough on a few small problems to be worth further investigation. The Forward-Forward algorithm replaces the forward and backward passes of backpropagation by two forward passes, one with positive (i.e. real) data and the other with negative data which could be generated by the network itself. Each layer has its own objective function which is simply to have high goodness for positive data and low goodness for negative data. The sum of the squared activities in a layer can be used as the goodness but there are many other possibilities, including minus the sum of the squared activities. If the positive and negative passes could be separated in time, the negative passes could be done offline, which would make the learning much simpler in the positive pass and allow video to be pipelined through the network without ever storing activities or stopping to propagate derivatives.