Equivalence of equilibrium propagation and recurrent backpropagation

DOI10.1162/NECO_A_01160zbMATH Open1474.68271arXiv1711.08416OpenAlexW2963953025WikidataQ90701059 ScholiaQ90701059MaRDI QIDQ3379592FDOQ3379592

Authors: Benjamin Scellier, Yoshua Bengio 

Publication date: 27 September 2021

Published in: Neural Computation (Search for Journal in Brave)

Abstract: Recurrent Backpropagation and Equilibrium Propagation are supervised learning algorithms for fixed point recurrent neural networks which differ in their second phase. In the first phase, both algorithms converge to a fixed point which corresponds to the configuration where the prediction is made. In the second phase, Equilibrium Propagation relaxes to another nearby fixed point corresponding to smaller prediction error, whereas Recurrent Backpropagation uses a side network to compute error derivatives iteratively. In this work we establish a close connection between these two algorithms. We show that, at every moment in the second phase, the temporal derivatives of the neural activities in Equilibrium Propagation are equal to the error derivatives computed iteratively by Recurrent Backpropagation in the side network. This work shows that it is not required to have a side network for the computation of error derivatives, and supports the hypothesis that, in biological neural networks, temporal derivatives of neural activities may code for error signals.

Full work available at URL: https://arxiv.org/abs/1711.08416

Recommendations

• Equivalence of Backpropagation and Contrastive Hebbian Learning in a Layered Network
• Learning fixed point patterns by recurrent networks
• Predictive coding approximates backprop along arbitrary computation graphs
• Timescale separation in recurrent neural networks
• STDP-compatible approximation of backpropagation in an energy-based model

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05) Neural networks for/in biological studies, artificial life and related topics (92B20)

Cites Work

• Neurons with graded response have collective computational properties like those of two-state neurons
• Absolute stability of global pattern formation and parallel memory storage by competitive neural networks

Cited In (5)

• Equivalence of Backpropagation and Contrastive Hebbian Learning in a Layered Network
• STDP-compatible approximation of backpropagation in an energy-based model
• Predictive coding approximates backprop along arbitrary computation graphs
• Flexible working memory through selective gating and attentional tagging
• Supervised learning through physical changes in a mechanical system

This page was built for publication: Equivalence of equilibrium propagation and recurrent backpropagation

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3379592)