Trainable hardware for dynamical computing using error backpropagation through physical media.
Identifieur interne : 000098 ( PubMed/Corpus ); précédent : 000097; suivant : 000099Trainable hardware for dynamical computing using error backpropagation through physical media.
Auteurs : Michiel Hermans ; Michaël Burm ; Thomas Van Vaerenbergh ; Joni Dambre ; Peter BienstmanSource :
- Nature communications ; 2015.
Abstract
Neural networks are currently implemented on digital Von Neumann machines, which do not fully leverage their intrinsic parallelism. We demonstrate how to use a novel class of reconfigurable dynamical systems for analogue information processing, mitigating this problem. Our generic hardware platform for dynamic, analogue computing consists of a reciprocal linear dynamical system with nonlinear feedback. Thanks to reciprocity, a ubiquitous property of many physical phenomena like the propagation of light and sound, the error backpropagation-a crucial step for tuning such systems towards a specific task-can happen in hardware. This can potentially speed up the optimization process significantly, offering important benefits for the scalability of neuro-inspired hardware. In this paper, we show, using one experimentally validated and one conceptual example, that such systems may provide a straightforward mechanism for constructing highly scalable, fully dynamical analogue computers.
DOI: 10.1038/ncomms7729
PubMed: 25801303
Links to Exploration step
pubmed:25801303Le document en format XML
<record><TEI><teiHeader><fileDesc><titleStmt><title xml:lang="en">Trainable hardware for dynamical computing using error backpropagation through physical media.</title><author><name sortKey="Hermans, Michiel" sort="Hermans, Michiel" uniqKey="Hermans M" first="Michiel" last="Hermans">Michiel Hermans</name><affiliation><nlm:affiliation>OPERA photonique, Université Libre de Bruxelles, Avenue F. Roosevelt 50, 1050 Brussels, Belgium.</nlm:affiliation></affiliation></author><author><name sortKey="Burm, Michael" sort="Burm, Michael" uniqKey="Burm M" first="Michaël" last="Burm">Michaël Burm</name><affiliation><nlm:affiliation>ELIS Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</nlm:affiliation></affiliation></author><author><name sortKey="Van Vaerenbergh, Thomas" sort="Van Vaerenbergh, Thomas" uniqKey="Van Vaerenbergh T" first="Thomas" last="Van Vaerenbergh">Thomas Van Vaerenbergh</name><affiliation><nlm:affiliation>INTEC Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</nlm:affiliation></affiliation></author><author><name sortKey="Dambre, Joni" sort="Dambre, Joni" uniqKey="Dambre J" first="Joni" last="Dambre">Joni Dambre</name><affiliation><nlm:affiliation>ELIS Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</nlm:affiliation></affiliation></author><author><name sortKey="Bienstman, Peter" sort="Bienstman, Peter" uniqKey="Bienstman P" first="Peter" last="Bienstman">Peter Bienstman</name><affiliation><nlm:affiliation>INTEC Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</nlm:affiliation></affiliation></author></titleStmt><publicationStmt><idno type="wicri:source">PubMed</idno><date when="2015">2015</date><idno type="doi">10.1038/ncomms7729</idno><idno type="RBID">pubmed:25801303</idno><idno type="pmid">25801303</idno><idno type="wicri:Area/PubMed/Corpus">000098</idno></publicationStmt><sourceDesc><biblStruct><analytic><title xml:lang="en">Trainable hardware for dynamical computing using error backpropagation through physical media.</title><author><name sortKey="Hermans, Michiel" sort="Hermans, Michiel" uniqKey="Hermans M" first="Michiel" last="Hermans">Michiel Hermans</name><affiliation><nlm:affiliation>OPERA photonique, Université Libre de Bruxelles, Avenue F. Roosevelt 50, 1050 Brussels, Belgium.</nlm:affiliation></affiliation></author><author><name sortKey="Burm, Michael" sort="Burm, Michael" uniqKey="Burm M" first="Michaël" last="Burm">Michaël Burm</name><affiliation><nlm:affiliation>ELIS Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</nlm:affiliation></affiliation></author><author><name sortKey="Van Vaerenbergh, Thomas" sort="Van Vaerenbergh, Thomas" uniqKey="Van Vaerenbergh T" first="Thomas" last="Van Vaerenbergh">Thomas Van Vaerenbergh</name><affiliation><nlm:affiliation>INTEC Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</nlm:affiliation></affiliation></author><author><name sortKey="Dambre, Joni" sort="Dambre, Joni" uniqKey="Dambre J" first="Joni" last="Dambre">Joni Dambre</name><affiliation><nlm:affiliation>ELIS Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</nlm:affiliation></affiliation></author><author><name sortKey="Bienstman, Peter" sort="Bienstman, Peter" uniqKey="Bienstman P" first="Peter" last="Bienstman">Peter Bienstman</name><affiliation><nlm:affiliation>INTEC Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</nlm:affiliation></affiliation></author></analytic><series><title level="j">Nature communications</title><idno type="e-ISSN">2041-1723</idno><imprint><date when="2015" type="published">2015</date></imprint></series></biblStruct></sourceDesc></fileDesc><profileDesc><textClass></textClass></profileDesc></teiHeader><front><div type="abstract" xml:lang="en">Neural networks are currently implemented on digital Von Neumann machines, which do not fully leverage their intrinsic parallelism. We demonstrate how to use a novel class of reconfigurable dynamical systems for analogue information processing, mitigating this problem. Our generic hardware platform for dynamic, analogue computing consists of a reciprocal linear dynamical system with nonlinear feedback. Thanks to reciprocity, a ubiquitous property of many physical phenomena like the propagation of light and sound, the error backpropagation-a crucial step for tuning such systems towards a specific task-can happen in hardware. This can potentially speed up the optimization process significantly, offering important benefits for the scalability of neuro-inspired hardware. In this paper, we show, using one experimentally validated and one conceptual example, that such systems may provide a straightforward mechanism for constructing highly scalable, fully dynamical analogue computers.</div></front></TEI><pubmed><MedlineCitation Owner="NLM" Status="PubMed-not-MEDLINE"><PMID Version="1">25801303</PMID><DateCreated><Year>2015</Year><Month>03</Month><Day>24</Day></DateCreated><DateCompleted><Year>2015</Year><Month>08</Month><Day>18</Day></DateCompleted><DateRevised><Year>2015</Year><Month>04</Month><Day>08</Day></DateRevised><Article PubModel="Electronic"><Journal><ISSN IssnType="Electronic">2041-1723</ISSN><JournalIssue CitedMedium="Internet"><Volume>6</Volume><PubDate><Year>2015</Year></PubDate></JournalIssue><Title>Nature communications</Title><ISOAbbreviation>Nat Commun</ISOAbbreviation></Journal><ArticleTitle>Trainable hardware for dynamical computing using error backpropagation through physical media.</ArticleTitle><Pagination><MedlinePgn>6729</MedlinePgn></Pagination><ELocationID EIdType="doi" ValidYN="Y">10.1038/ncomms7729</ELocationID><Abstract><AbstractText>Neural networks are currently implemented on digital Von Neumann machines, which do not fully leverage their intrinsic parallelism. We demonstrate how to use a novel class of reconfigurable dynamical systems for analogue information processing, mitigating this problem. Our generic hardware platform for dynamic, analogue computing consists of a reciprocal linear dynamical system with nonlinear feedback. Thanks to reciprocity, a ubiquitous property of many physical phenomena like the propagation of light and sound, the error backpropagation-a crucial step for tuning such systems towards a specific task-can happen in hardware. This can potentially speed up the optimization process significantly, offering important benefits for the scalability of neuro-inspired hardware. In this paper, we show, using one experimentally validated and one conceptual example, that such systems may provide a straightforward mechanism for constructing highly scalable, fully dynamical analogue computers.</AbstractText></Abstract><AuthorList CompleteYN="Y"><Author ValidYN="Y"><LastName>Hermans</LastName><ForeName>Michiel</ForeName><Initials>M</Initials><AffiliationInfo><Affiliation>OPERA photonique, Université Libre de Bruxelles, Avenue F. Roosevelt 50, 1050 Brussels, Belgium.</Affiliation></AffiliationInfo></Author><Author ValidYN="Y"><LastName>Burm</LastName><ForeName>Michaël</ForeName><Initials>M</Initials><AffiliationInfo><Affiliation>ELIS Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</Affiliation></AffiliationInfo></Author><Author ValidYN="Y"><LastName>Van Vaerenbergh</LastName><ForeName>Thomas</ForeName><Initials>T</Initials><AffiliationInfo><Affiliation>INTEC Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</Affiliation></AffiliationInfo></Author><Author ValidYN="Y"><LastName>Dambre</LastName><ForeName>Joni</ForeName><Initials>J</Initials><AffiliationInfo><Affiliation>ELIS Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</Affiliation></AffiliationInfo></Author><Author ValidYN="Y"><LastName>Bienstman</LastName><ForeName>Peter</ForeName><Initials>P</Initials><AffiliationInfo><Affiliation>INTEC Department, Ghent University, Sint Pietersnieuwstraat 41, 9000 Ghent, Belgium.</Affiliation></AffiliationInfo></Author></AuthorList><Language>eng</Language><PublicationTypeList><PublicationType UI="D016428">Journal Article</PublicationType><PublicationType UI="D013485">Research Support, Non-U.S. Gov't</PublicationType></PublicationTypeList><ArticleDate DateType="Electronic"><Year>2015</Year><Month>03</Month><Day>24</Day></ArticleDate></Article><MedlineJournalInfo><Country>England</Country><MedlineTA>Nat Commun</MedlineTA><NlmUniqueID>101528555</NlmUniqueID><ISSNLinking>2041-1723</ISSNLinking></MedlineJournalInfo><CommentsCorrectionsList><CommentsCorrections RefType="Cites"><RefSource>Artif Life. 2013 Winter;19(1):35-66</RefSource><PMID Version="1">23186351</PMID></CommentsCorrections><CommentsCorrections RefType="Cites"><RefSource>IEEE Trans Neural Netw Learn Syst. 2015 Jul;26(7):1545-50</RefSource><PMID Version="1">25137733</PMID></CommentsCorrections><CommentsCorrections RefType="Cites"><RefSource>Neural Comput. 2002 Nov;14(11):2531-60</RefSource><PMID Version="1">12433288</PMID></CommentsCorrections><CommentsCorrections RefType="Cites"><RefSource>Opt Express. 2008 Jul 21;16(15):11182-92</RefSource><PMID Version="1">18648434</PMID></CommentsCorrections><CommentsCorrections RefType="Cites"><RefSource>Opt Express. 2009 Dec 21;17(26):24020-9</RefSource><PMID Version="1">20052114</PMID></CommentsCorrections><CommentsCorrections RefType="Cites"><RefSource>Opt Express. 2012 Jan 30;20(3):3241-9</RefSource><PMID Version="1">22330562</PMID></CommentsCorrections><CommentsCorrections RefType="Cites"><RefSource>PLoS One. 2014;9(1):e86696</RefSource><PMID Version="1">24497969</PMID></CommentsCorrections></CommentsCorrectionsList><OtherID Source="NLM">PMC4382991</OtherID></MedlineCitation><PubmedData><History><PubMedPubDate PubStatus="received"><Year>2014</Year><Month>12</Month><Day>22</Day></PubMedPubDate><PubMedPubDate PubStatus="accepted"><Year>2015</Year><Month>2</Month><Day>23</Day></PubMedPubDate><PubMedPubDate PubStatus="entrez"><Year>2015</Year><Month>3</Month><Day>25</Day><Hour>6</Hour><Minute>0</Minute></PubMedPubDate><PubMedPubDate PubStatus="pubmed"><Year>2015</Year><Month>3</Month><Day>25</Day><Hour>6</Hour><Minute>0</Minute></PubMedPubDate><PubMedPubDate PubStatus="medline"><Year>2015</Year><Month>3</Month><Day>25</Day><Hour>6</Hour><Minute>1</Minute></PubMedPubDate></History><PublicationStatus>epublish</PublicationStatus><ArticleIdList><ArticleId IdType="pii">ncomms7729</ArticleId><ArticleId IdType="doi">10.1038/ncomms7729</ArticleId><ArticleId IdType="pubmed">25801303</ArticleId><ArticleId IdType="pmc">PMC4382991</ArticleId></ArticleIdList></PubmedData></pubmed></record>Pour manipuler ce document sous Unix (Dilib)
EXPLOR_STEP=$WICRI_ROOT/Wicri/Musique/explor/OperaV1/Data/PubMed/Corpus
HfdSelect -h $EXPLOR_STEP/biblio.hfd -nk 000098 | SxmlIndent | more
Ou
HfdSelect -h $EXPLOR_AREA/Data/PubMed/Corpus/biblio.hfd -nk 000098 | SxmlIndent | more
Pour mettre un lien sur cette page dans le réseau Wicri
{{Explor lien
|wiki= Wicri/Musique
|area= OperaV1
|flux= PubMed
|étape= Corpus
|type= RBID
|clé= pubmed:25801303
|texte= Trainable hardware for dynamical computing using error backpropagation through physical media.
}}
Pour générer des pages wiki
HfdIndexSelect -h $EXPLOR_AREA/Data/PubMed/Corpus/RBID.i -Sk "pubmed:25801303" \
| HfdSelect -Kh $EXPLOR_AREA/Data/PubMed/Corpus/biblio.hfd \
| NlmPubMed2Wicri -a OperaV1
| This area was generated with Dilib version V0.6.21. |  |