Toward Scalable, Efficient, and Accurate Deep Spiking Neural Networks With Backward Residual Connections, Stochastic Softmax, and Hybridization
Spiking Neural Networks (SNNs) may offer an energy-efficient alternative for implementing deep learning applications. In recent years, there have been several proposals focused on supervised (conversion, spike-based gradient descent) and unsupervised (spike timing dependent plasticity) training meth...
Main Authors: | , , |
---|---|
Format: | Article |
Language: | English |
Published: |
Frontiers Media S.A.
2020-06-01
|
Series: | Frontiers in Neuroscience |
Subjects: | |
Online Access: | https://www.frontiersin.org/article/10.3389/fnins.2020.00653/full |
_version_ | 1811310353267556352 |
---|---|
author | Priyadarshini Panda Sai Aparna Aketi Kaushik Roy |
author_facet | Priyadarshini Panda Sai Aparna Aketi Kaushik Roy |
author_sort | Priyadarshini Panda |
collection | DOAJ |
description | Spiking Neural Networks (SNNs) may offer an energy-efficient alternative for implementing deep learning applications. In recent years, there have been several proposals focused on supervised (conversion, spike-based gradient descent) and unsupervised (spike timing dependent plasticity) training methods to improve the accuracy of SNNs on large-scale tasks. However, each of these methods suffer from scalability, latency, and accuracy limitations. In this paper, we propose novel algorithmic techniques of modifying the SNN configuration with backward residual connections, stochastic softmax, and hybrid artificial-and-spiking neuronal activations to improve the learning ability of the training methodologies to yield competitive accuracy, while, yielding large efficiency gains over their artificial counterparts. Note, artificial counterparts refer to conventional deep learning/artificial neural networks. Our techniques apply to VGG/Residual architectures, and are compatible with all forms of training methodologies. Our analysis reveals that the proposed solutions yield near state-of-the-art accuracy with significant energy-efficiency and reduced parameter overhead translating to hardware improvements on complex visual recognition tasks, such as, CIFAR10, Imagenet datatsets. |
first_indexed | 2024-04-13T09:57:29Z |
format | Article |
id | doaj.art-54e29c8a75424bef866d89cce34ed123 |
institution | Directory Open Access Journal |
issn | 1662-453X |
language | English |
last_indexed | 2024-04-13T09:57:29Z |
publishDate | 2020-06-01 |
publisher | Frontiers Media S.A. |
record_format | Article |
series | Frontiers in Neuroscience |
spelling | doaj.art-54e29c8a75424bef866d89cce34ed1232022-12-22T02:51:19ZengFrontiers Media S.A.Frontiers in Neuroscience1662-453X2020-06-011410.3389/fnins.2020.00653535502Toward Scalable, Efficient, and Accurate Deep Spiking Neural Networks With Backward Residual Connections, Stochastic Softmax, and HybridizationPriyadarshini Panda0Sai Aparna Aketi1Kaushik Roy2Department of Electrical Engineering, Yale University, New Haven, CT, United StatesSchool of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, United StatesSchool of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, United StatesSpiking Neural Networks (SNNs) may offer an energy-efficient alternative for implementing deep learning applications. In recent years, there have been several proposals focused on supervised (conversion, spike-based gradient descent) and unsupervised (spike timing dependent plasticity) training methods to improve the accuracy of SNNs on large-scale tasks. However, each of these methods suffer from scalability, latency, and accuracy limitations. In this paper, we propose novel algorithmic techniques of modifying the SNN configuration with backward residual connections, stochastic softmax, and hybrid artificial-and-spiking neuronal activations to improve the learning ability of the training methodologies to yield competitive accuracy, while, yielding large efficiency gains over their artificial counterparts. Note, artificial counterparts refer to conventional deep learning/artificial neural networks. Our techniques apply to VGG/Residual architectures, and are compatible with all forms of training methodologies. Our analysis reveals that the proposed solutions yield near state-of-the-art accuracy with significant energy-efficiency and reduced parameter overhead translating to hardware improvements on complex visual recognition tasks, such as, CIFAR10, Imagenet datatsets.https://www.frontiersin.org/article/10.3389/fnins.2020.00653/fullspiking neural networksenergy-efficiencybackward residual connectionstochastic softmaxhybridizationimproved accuracy |
spellingShingle | Priyadarshini Panda Sai Aparna Aketi Kaushik Roy Toward Scalable, Efficient, and Accurate Deep Spiking Neural Networks With Backward Residual Connections, Stochastic Softmax, and Hybridization Frontiers in Neuroscience spiking neural networks energy-efficiency backward residual connection stochastic softmax hybridization improved accuracy |
title | Toward Scalable, Efficient, and Accurate Deep Spiking Neural Networks With Backward Residual Connections, Stochastic Softmax, and Hybridization |
title_full | Toward Scalable, Efficient, and Accurate Deep Spiking Neural Networks With Backward Residual Connections, Stochastic Softmax, and Hybridization |
title_fullStr | Toward Scalable, Efficient, and Accurate Deep Spiking Neural Networks With Backward Residual Connections, Stochastic Softmax, and Hybridization |
title_full_unstemmed | Toward Scalable, Efficient, and Accurate Deep Spiking Neural Networks With Backward Residual Connections, Stochastic Softmax, and Hybridization |
title_short | Toward Scalable, Efficient, and Accurate Deep Spiking Neural Networks With Backward Residual Connections, Stochastic Softmax, and Hybridization |
title_sort | toward scalable efficient and accurate deep spiking neural networks with backward residual connections stochastic softmax and hybridization |
topic | spiking neural networks energy-efficiency backward residual connection stochastic softmax hybridization improved accuracy |
url | https://www.frontiersin.org/article/10.3389/fnins.2020.00653/full |
work_keys_str_mv | AT priyadarshinipanda towardscalableefficientandaccuratedeepspikingneuralnetworkswithbackwardresidualconnectionsstochasticsoftmaxandhybridization AT saiaparnaaketi towardscalableefficientandaccuratedeepspikingneuralnetworkswithbackwardresidualconnectionsstochasticsoftmaxandhybridization AT kaushikroy towardscalableefficientandaccuratedeepspikingneuralnetworkswithbackwardresidualconnectionsstochasticsoftmaxandhybridization |