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
Introducing GA-SSNN: A Method for Optimizing Stochastic Spiking Neural Networks: Scaling the Edge User Allocation Constraint Satisfaction Problem with Enhanced Energy and Time Efficiency
KTH, School of Electrical Engineering and Computer Science (EECS).
2023 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
Introducerar GA-SSNN: En metod för att optimera stokastiskt spikande neurala nätverk : Uppskalning av edge user allokering med förbättrad energi- och tidseffektivitet (Swedish)
Abstract [en]

As progress within Von Neumann-based computer architecture is being limited by the physical limits of transistor size, neuromorphic comuting has emerged as a promising area of research. Neuromorphic hardware tends to be substantially more power efficient by imitating the aspects of computations in networks of neurons in the brain. It features massive parallelism, colocation of processing and memory at the neurons and synapses, inherent scalability, temporally sparse event-driven computation, and stochasticity. This thesis explores the application of neuromorphic comuting, specifically Stochastic Spiking Neural Networks (SSNNs), to large-scale edge user allocation constraint satisfaction problems (CSPs). These problems are central in the era of 5G networks, augmented reality and computational offloading, yet existing solutions struggle with scalability and stability. The thesis introduces the Genetic Algorithm for Stochastic Spiking Neural Networks (GA-SSNN), an algorithm designed to optimize complex and stochastic objective functions. The GA-SSNN algorithm leverages adaptive mutation, simulation time management, constraint approximation, and specialized tournament selection to efficiently traverse the search space and achieves better performance than the current state of the art (NSGA-II). Furthermore, the thesis elaborates on designing an SSNN structure to efficiently solve a complex CSP. The outcome of this thesis represents a significant step towards applying neuromorphic computing to real-world scenarios, with the potential to greatly enhance solution speed and energy efficiency.

Abstract [sv]

Då utveckling inom Von Neumann-baserad datorarkitektur begränsas av de fysiska gränserna för transistorstorlek har neuromorphic computing blivit ett intressant forskningsområde. Neuromorfisk hårdvara är generellt mycket energieffektiv eftersom den imiterar hjärnans neuroner. Detta medför även massiv parallellism, att minne och beräkning är tätt sammankopplade, naturlig skalbarhet, händelsedriven beräkning och stokasticitet. Detta examensarbete utforskar tillämpning av neromorphic computing, i form av Stokastiskt Spikande Neurala Nätverk (SSNNs), på storskaliga edge userallocation constraint satisfaction problem. Dessa problem är centrala inom utbyggnaden av 5G-nätverket, augmented reality och inom computational offloading. Samtidigt lider nuvarande metoder för att lösa dem av skalningsoch stabilitetsproblem. Examensarbetet introducerar Genetic Algorithm for Stochastic Spiking Neural Networks (GA-SSNN), som är designad för att optimera komplexa och stokastiska “objective functions”. GASSNN-algoritmen använder sig av adaptiv mutation, adaptiv simulationstid, contraint approximation och specialiserad tournament selection för att effektivt utforska lösningsrymden och uppnår bättre prestanda än de tidigare bästa kända algoritmerna (NSGA-II). Vidare presenteras hur man designar en SSNN-struktur för att effektivt lösa komplexa CSP. Resultatet av detta examensarbete innebär en påtaglig förbättring i den potentiella användbarheten av neuromorphic computing, med potential att signifikant förbättra lösningstid och energieffektivitet.

Place, publisher, year, edition, pages
2023. , p. 63
Series
TRITA-EECS-EX ; 2023:878
National Category
Computer and Information Sciences
Identifiers
URN: urn:nbn:se:kth:diva-343445OAI: oai:DiVA.org:kth-343445DiVA, id: diva2:1837710
External cooperation
Ericsson AB
Supervisors
Examiners
Available from: 2024-02-15 Created: 2024-02-14 Last updated: 2025-02-14Bibliographically approved

Open Access in DiVA

fulltext(2465 kB)23 downloads
File information
File name FULLTEXT03.pdfFile size 2465 kBChecksum SHA-512
6b220a9ac7e72d47fb82b4e74c04a5c982951b6405f19ccfaa0680584ad3ae5267761466e78422ada74ff97316f24895133240d5375867d9d1e3aa68c99ac1ce
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: 175 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: 748 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