Digitala Vetenskapliga Arkivet

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
Evaluating the Quality of Dimensionality Reduction Techniques on Node and Graph Embeddings: Visualizing Graph Neural Networks
KTH, School of Electrical Engineering and Computer Science (EECS).
2024 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
Evaluating the Quality of Dimensionality Reduction Techniques on Node and Graph Embeddings (Swedish)
Abstract [en]

Graph datasets are increasingly prevalent in machine learning applications. Effective visualization is crucial for understanding both the datasets and the behavior of graph neural networks (GNNs) applied to them. Visualizing these complex structures remains a significant challenge. This thesis introduces a framework designed to overcome these challenges by visually analyzing graph datasets and the inner workings of GNNs. The visualizations are created through training graph neural networks on graph datasets, extracting layer embeddings, and applying dimensionality reduction techniques to visualize the data in a two-dimensional space. This method was applied to a variety of datasets, graph neural network architectures, and dimensionality reduction techniques. The visualizations are compared using cluster-quality metrics that quantify the intra-clustering and inter-clustering quality. The analysis shows that SplineCNN [1] network embeddings are the most easily separable, and UMAP [2] creates the best visual representation of the data.

Abstract [sv]

Graphdatamängder blir allt vanligare i maskininlärningsapplikationer. Effektiv visualisering är avgörande för att förstå både datamängderna och beteendet hos grafneurala nätverk (GNN) som tillämpas på dem. Visualiseringen av dessa komplexa strukturer är fortfarande en betydande utmaning. Denna avhandling introducerar ett ramverk utformat för att övervinna dessa utmaningar genom att visuellt analysera grafdatamängder och de inre mekanismerna hos GNN. Visualiseringarna skapas genom att träna grafneurala nätverk på grafdatamängder, extrahera lagers inbäddningar och tillämpa tekniker för dimensionsreduktion för att visualisera data i ett tvådimensionellt utrymme. Denna metod tillämpades på en mängd olika datamängder, grafneurala nätverksarkitekturer och tekniker för dimensionsreduktion. Visualiseringarna jämförs med hjälp av klusterkvalitetsmått som kvantifierar kvaliteten på intra-klustring och inter-klustring. Analysen visar att SplineCNN [1] nätverksinbäddningar är de mest lättseparerbara, och UMAP [2] skapar den bästa visuella representationen av data.

Place, publisher, year, edition, pages
2024. , p. 56
Series
TRITA-EECS-EX ; 2024:936
Keywords [en]
dimensionality reduction, visualization, graph neural networks, node embeddings, graph datasets
National Category
Computer and Information Sciences
Identifiers
URN: urn:nbn:se:kth:diva-361062OAI: oai:DiVA.org:kth-361062DiVA, id: diva2:1943599
External cooperation
Stanford Medicine
Supervisors
Examiners
Available from: 2025-03-17 Created: 2025-03-11 Last updated: 2025-03-17Bibliographically approved

Open Access in DiVA

fulltext(11123 kB)172 downloads
File information
File name FULLTEXT02.pdfFile size 11123 kBChecksum SHA-512
ca002ca85cfa99b8529f77c79007cf1ee8ec5ddd24b945707f563cc8a504dd881760d98a9ddddaf68e860fa71eb799b90dd9f7407a11bf9bb192d2ec3e9526d9
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: 172 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: 297 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