http://www.diva-portal.org/smash/record.jsf?pid=diva2%3A1457863&dswid=-740

DiVA

diva-portal.org

Digitala Vetenskapliga Arkivet

[ajax-loade] Please wait ...

Simple search

Advanced search -
Research publicationsAdvanced search -
Student thesesStatistics

EnglishSvenskaNorsk
Jump to content
Change search
[                                                            ]Search
Search [ ]Only documents with full text in DiVA
CiteExport

  * BibTex
  * CSL-JSON
  * CSV 1
  * CSV 2
  * CSV 3
  * CSV 4
  * CSV 5
  * CSV all metadata
  * CSV all metadata version 2
  * RIS
  * Mods
  * MARC-XML
  * ETDMS

Link to record
Permanent link
[https://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-168054    ] 
[                                                            ] 
[                                                            ] 

Direct link
[http://www.diva-portal.org/smash/record.jsf?pid=diva2:145786] 
[                                                            ] 
[                                                            ] 
Cite
Citation style
[                    ]
[apa                                    ]
 

  * apa
  * ieee
  * modern-language-association-8th-edition
  * vancouver
  * Other style

 
[                              ]More styles
Language
[                    ]
[en-GB       ]
 

  * de-DE
  * en-GB
  * en-US
  * fi-FI
  * nn-NO
  * nn-NB
  * sv-SE
  * Other locale

[          ]More languages
Output format
[                    ]
[html    ]
 

  * html
  * text
  * asciidoc
  * rtf

CreateClose
Evaluation of Machine Learning Primitives on a Digital Signal
Processor

Engstrom, Vilhelm

Linkoping University, Department of Science and Technology, Media and
Information Technology. Linkoping University, The Institute of
Technology.
2020 (English)Independent thesis Advanced level (degree of Master
(Two Years)), 20 credits / 30 HE creditsStudent thesis

Abstract [en]

Modern handheld devices rely on specialized hardware for evaluating
machine learning algorithms. This thesis investigates the feasibility
of using the digital signal processor, a part of the modem of the
device, as an alternative to this specialized hardware. Memory
management techniques and implementations for evaluating the machine
learning primitives convolutional, max-pooling and fully connected
layers are proposed. The implementations are evaluated based on to
what degree they utilize available hardware units. New instructions
for packing data and facilitating instruction pipelining are
suggested and evaluated. The results show that convolutional and
fully connected layers are well-suited to the processor used. The
aptness of the convolutional layer is subject to the kernel being
applied with a stride of 1 as larger strides cause the hardware usage
to plummet. Max-pooling layers, while not ill-suited, are the most
limited in terms of hardware usage. The proposed instructions are
shown to have positive effects on the throughput of the
implementations.

Place, publisher, year, edition, pages

2020. , p. 75

Keywords [en]

digital signal processor, DSP, SIMD, data parallelism, machine
learning, deep learning, convolutional neural network

National Category

Computer Engineering

Identifiers

URN: urn:nbn:se:liu:diva-168054ISRN: LiU-ITN-TEK-A--20/029--SEOAI:
oai:DiVA.org:liu-168054DiVA, id: diva2:1457863

Subject / course

Computer Engineering

Uppsok

Technology

Supervisors

Eilertsen, Gabriel

Linkoping University, Department of Science and Technology, Media and
Information Technology. Linkoping University, The Institute of
Technology.

Examiners

Nordman, Aida

Linkoping University, Department of Science and Technology, Media and
Information Technology. Linkoping University, The Institute of
Technology.
Available from: 2020-08-13 Created: 2020-08-13 Last updated: 
2020-08-13Bibliographically approved

Open Access in DiVA

 fulltext(635 kB)[info]286 downloads
[                                                            ] 
[                                                            ] 
[                                                            ] 

File information

File name FULLTEXT01.pdfFile size 635 kBChecksum SHA-512
d278ac8c5d88c631fa4e7cd17b9d8e7b484f568655551ac8073c4ebb8c57fcfe8e3579c6abb0760b5b2c8275403aa22595e464de1951e0375f204a8dc04f6e02
Type fulltextMimetype application/pdf

Search in DiVA

By author/editor

Engstrom, Vilhelm

By organisation

Media and Information TechnologyThe Institute of Technology

On the subject

Computer Engineering

Search outside of DiVA

GoogleGoogle Scholar
Total: 286 downloads[info]
The number of downloads is the sum of all downloads of full texts. It
may include eg previous versions that are now no longer available

[ajax-loade]

urn-nbn

Altmetric score

urn-nbn
Total: 422 hits
CiteExport

  * BibTex
  * CSL-JSON
  * CSV 1
  * CSV 2
  * CSV 3
  * CSV 4
  * CSV 5
  * CSV all metadata
  * CSV all metadata version 2
  * RIS
  * Mods
  * MARC-XML
  * ETDMS

Link to record
Permanent link
[https://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-168054    ] 
[                                                            ] 
[                                                            ] 

Direct link
[http://www.diva-portal.org/smash/record.jsf?pid=diva2:145786] 
[                                                            ] 
[                                                            ] 
Cite
Citation style
[                    ]
[apa                                    ]
 

  * apa
  * ieee
  * modern-language-association-8th-edition
  * vancouver
  * Other style

 
[                              ]More styles
Language
[                    ]
[en-GB       ]
 

  * de-DE
  * en-GB
  * en-US
  * fi-FI
  * nn-NO
  * nn-NB
  * sv-SE
  * Other locale

[          ]More languages
Output format
[                    ]
[html    ]
 

  * html
  * text
  * asciidoc
  * rtf

CreateClose
v. 2.43.0
|
WCAG
|
About DiVA Portal
|
About the DiVA Consortium