The backpropagation algorithm implemented on spiking neuromorphic hardware

Abstract The capabilities of natural neural systems have inspired both new generations of machine learning algorithms as well as neuromorphic, very large-scale integrated circuits capable of fast, low-power information processing. However, it has been argued that most modern machine learning algorit...

Full description

Saved in:
Bibliographic Details
Main Authors: Alpha Renner, Forrest Sheldon, Anatoly Zlotnik, Louis Tao, Andrew Sornborger
Format: Article
Language:English
Published: Nature Portfolio 2024-11-01
Series:Nature Communications
Online Access:https://doi.org/10.1038/s41467-024-53827-9
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850062268913418240
author Alpha Renner
Forrest Sheldon
Anatoly Zlotnik
Louis Tao
Andrew Sornborger
author_facet Alpha Renner
Forrest Sheldon
Anatoly Zlotnik
Louis Tao
Andrew Sornborger
author_sort Alpha Renner
collection DOAJ
description Abstract The capabilities of natural neural systems have inspired both new generations of machine learning algorithms as well as neuromorphic, very large-scale integrated circuits capable of fast, low-power information processing. However, it has been argued that most modern machine learning algorithms are not neurophysiologically plausible. In particular, the workhorse of modern deep learning, the backpropagation algorithm, has proven difficult to translate to neuromorphic hardware. This study presents a neuromorphic, spiking backpropagation algorithm based on synfire-gated dynamical information coordination and processing implemented on Intel’s Loihi neuromorphic research processor. We demonstrate a proof-of-principle three-layer circuit that learns to classify digits and clothing items from the MNIST and Fashion MNIST datasets. To our knowledge, this is the first work to show a Spiking Neural Network implementation of the exact backpropagation algorithm that is fully on-chip without a computer in the loop. It is competitive in accuracy with off-chip trained SNNs and achieves an energy-delay product suitable for edge computing. This implementation shows a path for using in-memory, massively parallel neuromorphic processors for low-power, low-latency implementation of modern deep learning applications.
format Article
id doaj-art-1bcbe702fd6249d98dfe0d1d908fbef5
institution DOAJ
issn 2041-1723
language English
publishDate 2024-11-01
publisher Nature Portfolio
record_format Article
series Nature Communications
spelling doaj-art-1bcbe702fd6249d98dfe0d1d908fbef52025-08-20T02:49:59ZengNature PortfolioNature Communications2041-17232024-11-0115111410.1038/s41467-024-53827-9The backpropagation algorithm implemented on spiking neuromorphic hardwareAlpha Renner0Forrest Sheldon1Anatoly Zlotnik2Louis Tao3Andrew Sornborger4Institute of Neuroinformatics, University of Zurich and ETH ZurichPhysics of Condensed Matter & Complex Systems (T-4), Los Alamos National LaboratoryApplied Mathematics & Plasma Physics (T-5), Los Alamos National LaboratoryCenter for Bioinformatics, National Laboratory of Protein Engineering and Plant Genetic Engineering, School of Life Sciences, Peking UniversityInformation Sciences (CCS-3), Los Alamos National LaboratoryAbstract The capabilities of natural neural systems have inspired both new generations of machine learning algorithms as well as neuromorphic, very large-scale integrated circuits capable of fast, low-power information processing. However, it has been argued that most modern machine learning algorithms are not neurophysiologically plausible. In particular, the workhorse of modern deep learning, the backpropagation algorithm, has proven difficult to translate to neuromorphic hardware. This study presents a neuromorphic, spiking backpropagation algorithm based on synfire-gated dynamical information coordination and processing implemented on Intel’s Loihi neuromorphic research processor. We demonstrate a proof-of-principle three-layer circuit that learns to classify digits and clothing items from the MNIST and Fashion MNIST datasets. To our knowledge, this is the first work to show a Spiking Neural Network implementation of the exact backpropagation algorithm that is fully on-chip without a computer in the loop. It is competitive in accuracy with off-chip trained SNNs and achieves an energy-delay product suitable for edge computing. This implementation shows a path for using in-memory, massively parallel neuromorphic processors for low-power, low-latency implementation of modern deep learning applications.https://doi.org/10.1038/s41467-024-53827-9
spellingShingle Alpha Renner
Forrest Sheldon
Anatoly Zlotnik
Louis Tao
Andrew Sornborger
The backpropagation algorithm implemented on spiking neuromorphic hardware
Nature Communications
title The backpropagation algorithm implemented on spiking neuromorphic hardware
title_full The backpropagation algorithm implemented on spiking neuromorphic hardware
title_fullStr The backpropagation algorithm implemented on spiking neuromorphic hardware
title_full_unstemmed The backpropagation algorithm implemented on spiking neuromorphic hardware
title_short The backpropagation algorithm implemented on spiking neuromorphic hardware
title_sort backpropagation algorithm implemented on spiking neuromorphic hardware
url https://doi.org/10.1038/s41467-024-53827-9
work_keys_str_mv AT alpharenner thebackpropagationalgorithmimplementedonspikingneuromorphichardware
AT forrestsheldon thebackpropagationalgorithmimplementedonspikingneuromorphichardware
AT anatolyzlotnik thebackpropagationalgorithmimplementedonspikingneuromorphichardware
AT louistao thebackpropagationalgorithmimplementedonspikingneuromorphichardware
AT andrewsornborger thebackpropagationalgorithmimplementedonspikingneuromorphichardware
AT alpharenner backpropagationalgorithmimplementedonspikingneuromorphichardware
AT forrestsheldon backpropagationalgorithmimplementedonspikingneuromorphichardware
AT anatolyzlotnik backpropagationalgorithmimplementedonspikingneuromorphichardware
AT louistao backpropagationalgorithmimplementedonspikingneuromorphichardware
AT andrewsornborger backpropagationalgorithmimplementedonspikingneuromorphichardware