← All research spotlights

Research spotlight · Unit 6

Geoffrey Hinton

University of Toronto

Research field
Neural networks, learning algorithms, and how the brain might learn

Geoffrey Hinton was an author of the 1986 paper that made backpropagation known and has worked on learning in neural networks for five decades; in 2024 he shared the Nobel Prize in Physics with John Hopfield for the foundations of machine learning with artificial neural networks. In a 2022 paper, written while he worked at Google Brain, he asked whether a network can learn without a backward pass at all, and proposed the forward-forward algorithm, in which two forward passes, one on real and one on negative data, train every layer by its own measure of goodness. On the MNIST digits the method came close to backpropagation, while being slower and generalizing a little less well. The unit presents it after the derivation of backpropagation, as the question what might replace the backward pass, in the brain and in low-power hardware.

In the masterclass

Section 4, after the backpropagation algorithm and the 1986 paper, presents his forward-forward algorithm as the question whether the backward pass is necessary.

The research is presented in Unit 6: Neural Networks and Backpropagation, in a research spotlight box that explains the result at the level of the unit and links it to what you have just learned.

The paper

Hinton, G. (2022). The Forward-Forward Algorithm: Some Preliminary Investigations. arXiv preprint, arXiv:2212.13345. https://doi.org/10.48550/arXiv.2212.13345

The scientists named here are not contributors to the AI Masterclass, are not affiliated with it, and have not endorsed it. The program presents and explains their published research, with references to the original papers.