Emergent representations in networks trained with the Forward-Forward algorithm

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Tosato, Niccolò, Basile, Lorenzo, Ballarin, Emanuele, de Alteriis, Giuseppe, Cazzaniga, Alberto, Ansuini, Alessio
Natura: Preprint
Pubblicazione: 2023
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917972223721472
author Tosato, Niccolò
Basile, Lorenzo
Ballarin, Emanuele
de Alteriis, Giuseppe
Cazzaniga, Alberto
Ansuini, Alessio
author_facet Tosato, Niccolò
Basile, Lorenzo
Ballarin, Emanuele
de Alteriis, Giuseppe
Cazzaniga, Alberto
Ansuini, Alessio
contents The Backpropagation algorithm has often been criticised for its lack of biological realism. In an attempt to find a more biologically plausible alternative, the recently introduced Forward-Forward algorithm replaces the forward and backward passes of Backpropagation with two forward passes. In this work, we show that the internal representations obtained by the Forward-Forward algorithm can organise into category-specific ensembles exhibiting high sparsity -- composed of a low number of active units. This situation is reminiscent of what has been observed in cortical sensory areas, where neuronal ensembles are suggested to serve as the functional building blocks for perception and action. Interestingly, while this sparse pattern does not typically arise in models trained with standard Backpropagation, it can emerge in networks trained with Backpropagation on the same objective proposed for the Forward-Forward algorithm.
format Preprint
id arxiv_https___arxiv_org_abs_2305_18353
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Emergent representations in networks trained with the Forward-Forward algorithm
Tosato, Niccolò
Basile, Lorenzo
Ballarin, Emanuele
de Alteriis, Giuseppe
Cazzaniga, Alberto
Ansuini, Alessio
Neural and Evolutionary Computing
Machine Learning
The Backpropagation algorithm has often been criticised for its lack of biological realism. In an attempt to find a more biologically plausible alternative, the recently introduced Forward-Forward algorithm replaces the forward and backward passes of Backpropagation with two forward passes. In this work, we show that the internal representations obtained by the Forward-Forward algorithm can organise into category-specific ensembles exhibiting high sparsity -- composed of a low number of active units. This situation is reminiscent of what has been observed in cortical sensory areas, where neuronal ensembles are suggested to serve as the functional building blocks for perception and action. Interestingly, while this sparse pattern does not typically arise in models trained with standard Backpropagation, it can emerge in networks trained with Backpropagation on the same objective proposed for the Forward-Forward algorithm.
title Emergent representations in networks trained with the Forward-Forward algorithm
topic Neural and Evolutionary Computing
Machine Learning
url https://arxiv.org/abs/2305.18353