Compact yet efficient hardware architecture for multilayer-perceptron neural networks
Autor(a) principal: | |
---|---|
Data de Publicação: | 2011 |
Outros Autores: | , |
Tipo de documento: | Artigo |
Idioma: | eng |
Título da fonte: | Sba: Controle & Automação Sociedade Brasileira de Automatica |
Texto Completo: | http://old.scielo.br/scielo.php?script=sci_arttext&pid=S0103-17592011000600010 |
Resumo: | There are several neural network implementations using either software, hardware-based or a hardware/software co-design. This work proposes a hardware architecture to implement an artificial neural network (ANN), whose topology is the multilayer perceptron (MLP). In this paper, we explore the parallelism of neural networks and allow on-thefly changes of the number of inputs, number of layers and number of neurons per layer of the net. This reconfigurability characteristic permits that any application of ANNs may be implemented using the proposed hardware. In order to reduce the processing time that is spent in arithmetic computation, a real number is represented using a fraction of integers. In this way, the arithmetics is limited to integer operations, performed by fast combinational circuits. A simple state machine is required to control sums and products of fractions. Sigmoid is used as the activation function in the proposed implementation. It is approximated by polynomials, whose underlying computation requires only sums and products. A theorem is introduced and proven so as to cover the arithmetic strategy of the computation of the activation function. Thus, the arithmetic circuitry used to implement the neuron weighted sum is reused for computing the sigmoid. this resource sharing decreased drastically the total area of the system. After modeling and simulation for functionality validation, the proposed architecture synthesized using reconfigurable hardware. The results are promising. |
id |
SBA-2_bdced28c272040a92921bb3dbc769a7b |
---|---|
oai_identifier_str |
oai:scielo:S0103-17592011000600010 |
network_acronym_str |
SBA-2 |
network_name_str |
Sba: Controle & Automação Sociedade Brasileira de Automatica |
repository_id_str |
|
spelling |
Compact yet efficient hardware architecture for multilayer-perceptron neural networksArtificial neural networkshardware for neural networkssigmoidparallelismFPGAThere are several neural network implementations using either software, hardware-based or a hardware/software co-design. This work proposes a hardware architecture to implement an artificial neural network (ANN), whose topology is the multilayer perceptron (MLP). In this paper, we explore the parallelism of neural networks and allow on-thefly changes of the number of inputs, number of layers and number of neurons per layer of the net. This reconfigurability characteristic permits that any application of ANNs may be implemented using the proposed hardware. In order to reduce the processing time that is spent in arithmetic computation, a real number is represented using a fraction of integers. In this way, the arithmetics is limited to integer operations, performed by fast combinational circuits. A simple state machine is required to control sums and products of fractions. Sigmoid is used as the activation function in the proposed implementation. It is approximated by polynomials, whose underlying computation requires only sums and products. A theorem is introduced and proven so as to cover the arithmetic strategy of the computation of the activation function. Thus, the arithmetic circuitry used to implement the neuron weighted sum is reused for computing the sigmoid. this resource sharing decreased drastically the total area of the system. After modeling and simulation for functionality validation, the proposed architecture synthesized using reconfigurable hardware. The results are promising.Sociedade Brasileira de Automática2011-12-01info:eu-repo/semantics/articleinfo:eu-repo/semantics/publishedVersiontext/htmlhttp://old.scielo.br/scielo.php?script=sci_arttext&pid=S0103-17592011000600010Sba: Controle & Automação Sociedade Brasileira de Automatica v.22 n.6 2011reponame:Sba: Controle & Automação Sociedade Brasileira de Automaticainstname:Sociedade Brasileira de Automática (SBA)instacron:SBA10.1590/S0103-17592011000600010info:eu-repo/semantics/openAccessSilva,Rodrigo Martins daMourelle,Luiza de MacedoNedjah,Nadiaeng2012-01-13T00:00:00Zoai:scielo:S0103-17592011000600010Revistahttps://www.sba.org.br/revista/PUBhttps://old.scielo.br/oai/scielo-oai.php||revista_sba@fee.unicamp.br1807-03450103-1759opendoar:2012-01-13T00:00Sba: Controle & Automação Sociedade Brasileira de Automatica - Sociedade Brasileira de Automática (SBA)false |
dc.title.none.fl_str_mv |
Compact yet efficient hardware architecture for multilayer-perceptron neural networks |
title |
Compact yet efficient hardware architecture for multilayer-perceptron neural networks |
spellingShingle |
Compact yet efficient hardware architecture for multilayer-perceptron neural networks Silva,Rodrigo Martins da Artificial neural networks hardware for neural networks sigmoid parallelism FPGA |
title_short |
Compact yet efficient hardware architecture for multilayer-perceptron neural networks |
title_full |
Compact yet efficient hardware architecture for multilayer-perceptron neural networks |
title_fullStr |
Compact yet efficient hardware architecture for multilayer-perceptron neural networks |
title_full_unstemmed |
Compact yet efficient hardware architecture for multilayer-perceptron neural networks |
title_sort |
Compact yet efficient hardware architecture for multilayer-perceptron neural networks |
author |
Silva,Rodrigo Martins da |
author_facet |
Silva,Rodrigo Martins da Mourelle,Luiza de Macedo Nedjah,Nadia |
author_role |
author |
author2 |
Mourelle,Luiza de Macedo Nedjah,Nadia |
author2_role |
author author |
dc.contributor.author.fl_str_mv |
Silva,Rodrigo Martins da Mourelle,Luiza de Macedo Nedjah,Nadia |
dc.subject.por.fl_str_mv |
Artificial neural networks hardware for neural networks sigmoid parallelism FPGA |
topic |
Artificial neural networks hardware for neural networks sigmoid parallelism FPGA |
description |
There are several neural network implementations using either software, hardware-based or a hardware/software co-design. This work proposes a hardware architecture to implement an artificial neural network (ANN), whose topology is the multilayer perceptron (MLP). In this paper, we explore the parallelism of neural networks and allow on-thefly changes of the number of inputs, number of layers and number of neurons per layer of the net. This reconfigurability characteristic permits that any application of ANNs may be implemented using the proposed hardware. In order to reduce the processing time that is spent in arithmetic computation, a real number is represented using a fraction of integers. In this way, the arithmetics is limited to integer operations, performed by fast combinational circuits. A simple state machine is required to control sums and products of fractions. Sigmoid is used as the activation function in the proposed implementation. It is approximated by polynomials, whose underlying computation requires only sums and products. A theorem is introduced and proven so as to cover the arithmetic strategy of the computation of the activation function. Thus, the arithmetic circuitry used to implement the neuron weighted sum is reused for computing the sigmoid. this resource sharing decreased drastically the total area of the system. After modeling and simulation for functionality validation, the proposed architecture synthesized using reconfigurable hardware. The results are promising. |
publishDate |
2011 |
dc.date.none.fl_str_mv |
2011-12-01 |
dc.type.driver.fl_str_mv |
info:eu-repo/semantics/article |
dc.type.status.fl_str_mv |
info:eu-repo/semantics/publishedVersion |
format |
article |
status_str |
publishedVersion |
dc.identifier.uri.fl_str_mv |
http://old.scielo.br/scielo.php?script=sci_arttext&pid=S0103-17592011000600010 |
url |
http://old.scielo.br/scielo.php?script=sci_arttext&pid=S0103-17592011000600010 |
dc.language.iso.fl_str_mv |
eng |
language |
eng |
dc.relation.none.fl_str_mv |
10.1590/S0103-17592011000600010 |
dc.rights.driver.fl_str_mv |
info:eu-repo/semantics/openAccess |
eu_rights_str_mv |
openAccess |
dc.format.none.fl_str_mv |
text/html |
dc.publisher.none.fl_str_mv |
Sociedade Brasileira de Automática |
publisher.none.fl_str_mv |
Sociedade Brasileira de Automática |
dc.source.none.fl_str_mv |
Sba: Controle & Automação Sociedade Brasileira de Automatica v.22 n.6 2011 reponame:Sba: Controle & Automação Sociedade Brasileira de Automatica instname:Sociedade Brasileira de Automática (SBA) instacron:SBA |
instname_str |
Sociedade Brasileira de Automática (SBA) |
instacron_str |
SBA |
institution |
SBA |
reponame_str |
Sba: Controle & Automação Sociedade Brasileira de Automatica |
collection |
Sba: Controle & Automação Sociedade Brasileira de Automatica |
repository.name.fl_str_mv |
Sba: Controle & Automação Sociedade Brasileira de Automatica - Sociedade Brasileira de Automática (SBA) |
repository.mail.fl_str_mv |
||revista_sba@fee.unicamp.br |
_version_ |
1754824565647736832 |