Clustering the Nodes of Sparse Edge-weighted Graphs via Non-backtracking Spectra

Hannu Reittu, Marianna Bolla*, Fatma Abdelkhalek

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

Abstract

Theoretically supported techniques are given for clustering the nodes of edge-weighted graphs via non-backtracking spectra when the number of nodes is large and the skeleton graph is sparse. If the graph comes from a sparse stochastic block model, the structural real eigenvalues, out of the bulk of the spectrum, of the non-backtracking matrix are aligned with those of the expected adjacency matrix if it is of low rank. However, only the unweighted or weighted non-backtracking matrix is at our disposal. We show how the corresponding eigenvectors of the non-backtracking matrix and lower order companion matrices can be used to find assortative clusters of the nodes even in the case, when the expected adjacency matrix does not have a reduced rank, but it has a low-rank approximation. The paper gives the theoretical background and tools for sparse spectral clustering in very general frameworks. Application to sparse quantum chemistry networks is also presented.
Original languageEnglish
Title of host publicationAdvances in Information and Communication - Proceedings of the 2025 Future of Information and Communication Conference, FICC 2025
Subtitle of host publicationProceedings of the 2025 Future of Information and Communication Conference (FICC)
EditorsKohei Arai
PublisherSpringer
Pages81-99
Number of pages19
Volume2
ISBN (Electronic)978-3-031-85363-0
ISBN (Print)978-3-031-85362-3
DOIs
Publication statusPublished - 2025
MoE publication typeA4 Article in a conference publication
Event8th Future of Information and Communication Conference 2025 (FICC 2025) - Berlin, Germany
Duration: 28 Apr 202529 Apr 2025

Publication series

SeriesLecture Notes in Networks and Systems
Volume1284
ISSN2367-3370

Conference

Conference8th Future of Information and Communication Conference 2025 (FICC 2025)
Abbreviated titleFICC 2025
Country/TerritoryGermany
CityBerlin
Period28/04/2529/04/25

Keywords

  • Weighted non-backtracking matrix
  • Stochastic block models
  • K-means clustering

Fingerprint

Dive into the research topics of 'Clustering the Nodes of Sparse Edge-weighted Graphs via Non-backtracking Spectra'. Together they form a unique fingerprint.

Cite this