kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Comparison of Hebbian Learning and Backpropagation for Image Classification in Convolutional Neural Networks
KTH, School of Electrical Engineering and Computer Science (EECS).
KTH, School of Electrical Engineering and Computer Science (EECS).
2023 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
Abstract [en]

Current commonly used image recognition convolutional neural networks sharesome similarities with the human brain. However, the differences are many and the wellestablished backpropagation learning algorithm is not biologically plausible. Hebbianlearning is an algorithm that could minimize these differences and potentially provide imagerecognition networks with brain-like advantageous features. Here we explore the differencesbetween Hebbian learning and backpropagation, both regarding accuracy andrepresentations of data in hidden layers. Overall Hebbian networks performed considerablyworse than conventional backpropagation-trained networks. Experiments with incompletetraining data and distorted test data resulted in smaller but still visible performancedifferences. However, the convolutional filter structure of Hebbian networks proved to besimpler and more interpretable than the backpropagation equivalent. We hypothesize thatimprovements to increase scaling capabilities of Hebbian networks could render them apowerful alternative for image classification networks with more brain-like behavior.

Abstract [sv]

Convolutional neural networks, vanligt använda nätverk förbildigenkänning, är konstruerade för att efterlikna den mänskliga hjärnans struktur. Det finnsdock stora skillnader och den vanligaste inlärningsalgoritmen bakåtpropagering är inte rimligur ett biologiskt perspektiv. Hebbiansk inlärning är en algoritm som kan minska dessaskillnader och potentiellt ge bildigenkänningsnätverk fördelaktiga, hjärnlika egenskaper. Härutforskar vi skillnaderna mellan bakåtpropagering och hebbiansk inlärning, med avseende påträffsäkerhet och representationer av data i nätverkens dolda lager. Sammantagetpresterade hebbianska nätverk betydligt sämre än konventionella nätverk. Experiment medmindre träningsdata och förvrängd testdata resulterade i mindre men fortfarande synligaprestandaskillnader. Däremot visade sig strukturen för faltningsfiltren i hebbianska nätverkvara enklare och mer tolkningsbar än motsvarande filter för standardnätverken. Vi föreslåratt förbättringar för att öka skalningsmöjligheter för hebbianska nätverk kan göra dem till ettkraftfullt alternativ för dagens bildklassificeringsnätverk genom ett mer hjärnliknandebeteende.

Place, publisher, year, edition, pages
2023. , p. 673-679
Series
TRITA-EECS-EX ; 2023:196
National Category
Electrical Engineering, Electronic Engineering, Information Engineering
Identifiers
URN: urn:nbn:se:kth:diva-341786OAI: oai:DiVA.org:kth-341786DiVA, id: diva2:1823498
Supervisors
Examiners
Projects
Kandidatexjobb i elektroteknik 2023, KTH, StockholmAvailable from: 2024-01-02 Created: 2024-01-02

Open Access in DiVA

fulltext(211487 kB)436 downloads
File information
File name FULLTEXT01.pdfFile size 211487 kBChecksum SHA-512
69786101c351a58f7bd524c3aeee40c661028b577366c4a725033372b88c624c87c2183b6acca2d3d43bbd2bb2f3942326c69263e70c99cf1db027ce9c4e9ae2
Type fulltextMimetype application/pdf

By organisation
School of Electrical Engineering and Computer Science (EECS)
Electrical Engineering, Electronic Engineering, Information Engineering

Search outside of DiVA

GoogleGoogle Scholar
Total: 436 downloads
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

urn-nbn

Altmetric score

urn-nbn
Total: 191 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf