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
Adversarial robustness of STDP-trained spiking 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 thesisAlternative title
Robusthet av STDP-tränade spikande neuronnät mot fientliga attacker (Swedish)
Abstract [en]

Adversarial attacks on machine learning models are designed to elicit the wrong behavior from the model. One such attack on image classifiers are maliciously crafted inputs that, to the human eye, look untampered with but have been carefully altered to cause misclassification. Previous research has shown that spiking neural networks (SNN) trained with backpropagation can be more robust than, the more commonly used, artificial neural networks (ANN) against these attacks. In this thesis we conducted, to the best of our knowledge, novel research regarding adversarial attacks on SNNs trained with spike-timing-dependent plasticity (STDP), attacking the networks as well as analyzing their adversarial robustness compared to other neural networks. One of the reasons for attacking STDP-trained models is that STDP is more biologically plausible compared to other learning techniques for SNNs. The method used in this thesis is to implement multiple machine learning models based on different approaches and to compare their robustness with each other. The models consisted of two SNNs, trained with STDP and backpropagation through time (BPTT), and one ANN. The results shows that it is possible to fool STDP-trained SNNs with adversarial attacks and also indicates that the SNN trained with STDP is the most robust out of these networks.

Abstract [sv]

Fientliga attacker mot maskininlärningsmodeller är utformade för att framkalla felaktigt beteende från modellen. En sådan attack mot bildklassificerare är skapade indata som, för det mänskliga ögat, ser oförändrade ut men noggrant ändrats för att orsaka felklassificering. Tidigare forskning har visat att spikande neuronnät (SNN) tränade med backpropagation kan vara mer robusta än de, mer vanligt använda, artificiella neurala nätverken (ANN) mot sådana attacker. I denna rapport har vi genomfört, enligt vår kännedom, ny forskning om fientliga attacker mot SNN:er tränade med spike-timing-beroende plasticitet (STDP), där vi attackerade nätverken och analyserade deras robusthet jämfört med andra neurala nätverk. En av anledningarna till att attackera STDP-tränade modeller är att STDP är mer biologiskt rimligt jämfört med andra inlärningstekniker för SNN:er. Metoden som används i denna rapport är att implementera flera olika maskininlärningsmodeller och jämföra deras robusthet med varandra. Modellerna bestod av två SNN:er, tränade med STDP och backpropagation över tid (BPTT), samt ett ANN. Resultaten visar att det är möjligt att lura STDP-tränade SNN:er med fientliga attacker och indikerar också på att SNN:en tränad med STDP är den mest robusta av dessa nätverk.

Place, publisher, year, edition, pages
2023. , p. 28
Series
TRITA-EECS-EX ; 2023:266
National Category
Computer and Information Sciences
Identifiers
URN: urn:nbn:se:kth:diva-330753OAI: oai:DiVA.org:kth-330753DiVA, id: diva2:1778353
Supervisors
Examiners
Available from: 2023-07-27 Created: 2023-07-01 Last updated: 2023-07-27Bibliographically approved

Open Access in DiVA

fulltext(1385 kB)324 downloads
File information
File name FULLTEXT01.pdfFile size 1385 kBChecksum SHA-512
36b1e7cb860c3e32e7e3fdb1324fed3d072a7ec5610afbaa734f1a343bcbeb25e55479c7cc14c6ee6a1f0369df336a1f0b08b8effbb302c336c916d264056c81
Type fulltextMimetype application/pdf

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

Search outside of DiVA

GoogleGoogle Scholar
Total: 324 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: 496 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