Abstract

Geoffrey Hinton, presented the keynote during NeurIPS 2022 on his potential alternative to backpropagation; the Forward Forward Algorithm. This method proves closer to a neural model of the brain, and is a promising alternative to backpropagation in the case of power sentive hardware. On a small subset of toy problems, the algorithm is comparable to backpropagation, however, does not require neural activities to be stored.

The original work by Hinton can be found here.