Repository logo
 
Publication

A configurable architecture for running hybrid convolutional neural networks in low-density FPGAs

2020-06-08Journal article Open access
Simple item page
dc.contributor.authorVéstias, Mário
dc.contributor.authorDuarte, Rui
dc.contributor.authorDe Sousa, Jose
dc.contributor.authorCláudio de Campos Neto, Horácio
dc.date.accessioned2021-01-28T16:48:41Z
dc.date.available2021-01-28T16:48:41Z
dc.date.issued2020-06-08
dc.description.abstractConvolutional neural networks have become the state of the art of machine learning for a vast set of applications, especially for image classification and object detection. There are several advantages to running inference on these models at the edge, including real-time performance and data privacy. The high computing and memory requirements of convolutional neural networks have been major obstacles to the broader deployment of CNNs on edge devices. Data quantization is an optimization method that reduces the number of bits used to represent weights and activations of a network model, minimizing storage requirements and computing complexity. Quantization can be applied at the layer level, by using different bit widths in different layers: this is called hybrid quantization. This article proposes a new efficient and configurable architecture for running CNNs with hybrid quantization in low-density Field-Programmable Gate Arrays (FPGAs) targeting edge devices. The architecture has been implemented on the Xilinx ZYNQ7020/45 devices and is running the AlexNet and VGG16 networks. Running AlexNet, the architecture has a throughput up to 508 images per second on the ZYNQ7020 device, and 1639 images per second on the ZYNQ7045 device. Considering VGG16, the architecture delivers up to 43 images per second on the ZYNQ7020 device, and 81 images per second on the ZYNQ7045 device. The proposed hybrid architecture achieves up to 13.7 x improvement in performance compared to state-of-the-art solutions, with small accuracy degradation.pt_PT
dc.description.versioninfo:eu-repo/semantics/publishedVersionpt_PT
dc.identifier.citationVÉSTIAS, Mário P.; [et al] – A configurable architecture for running hybrid convolutional neural networks in low-density FPGAs. IEEE Access. ISSN 2169-3536. Vol. 8 (2020), pp. 107229-107243pt_PT
dc.identifier.doi10.1109/ACCESS.2020.3000444pt_PT
dc.identifier.issn2169-3536
dc.identifier.urihttp://hdl.handle.net/10400.21/12726
dc.language.isoengpt_PT
dc.peerreviewedyespt_PT
dc.publisherIEEEpt_PT
dc.relationUID/CEC/50021/2019 - FCTpt_PT
dc.relationUIDB/50021/2020 - FCTpt_PT
dc.relationPTDC/EEI-HAC/31819/2017 - project SARRROCA through INESC-IDpt_PT
dc.relation.publisherversionhttps://ieeexplore.ieee.org/ielx7/6287639/8948470/09110581.pdf?tag=1pt_PT
dc.subjectConvolutional neural networkpt_PT
dc.subjectDeep learningpt_PT
dc.subjectEmbedded computingpt_PT
dc.subjectField-programmable gate arraypt_PT
dc.subjectHybrid quantizationpt_PT
dc.titleA configurable architecture for running hybrid convolutional neural networks in low-density FPGAspt_PT
dc.typejournal article
dspace.entity.typePublication
oaire.citation.endPage107243pt_PT
oaire.citation.startPage107229pt_PT
oaire.citation.titleIEEE Accesspt_PT
oaire.citation.volume8pt_PT
person.familyNameVéstias
person.familyNameDuarte
person.familyNamede Sousa
person.familyNameCláudio de Campos Neto
person.givenNameMário
person.givenNameRui
person.givenNameJose
person.givenNameHorácio
person.identifier.ciencia-id4717-C2C7-3F2C
person.identifier.ciencia-idB91E-770F-19A3
person.identifier.ciencia-idBE18-E262-E0EC
person.identifier.ciencia-id9915-3BDF-5C35
person.identifier.orcid0000-0001-8556-4507
person.identifier.orcid0000-0002-7060-4745
person.identifier.orcid0000-0001-7525-7546
person.identifier.orcid0000-0002-3621-8322
person.identifier.ridH-9953-2012
person.identifier.ridI-4402-2015
person.identifier.ridL-6859-2015
person.identifier.scopus-author-id14525867300
person.identifier.scopus-author-id24823991600
person.identifier.scopus-author-id7102813024
rcaap.rightsopenAccesspt_PT
rcaap.typearticlept_PT
relation.isAuthorOfPublicationa7d22b29-c961-45ac-bc09-cd5e1002f1e8
relation.isAuthorOfPublicationf2b4b9e6-6c89-48c7-bc83-62d2e98a787b
relation.isAuthorOfPublicationd98a4d45-2d45-42ec-9f1d-14775723709b
relation.isAuthorOfPublication38334d5e-83e8-494c-a9e0-396299376d97
relation.isAuthorOfPublication.latestForDiscoveryf2b4b9e6-6c89-48c7-bc83-62d2e98a787b

Files

Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
A configurable_MPVestias.pdf
Size:
1.68 MB
Format:
Adobe Portable Document Format
Download

Collections

ISEL - Eng. Elect. Tel. Comp. - Artigos