Taştan, Aylin ; Muma, Michael ; Ollila, Esa ; Zoubir, Abdelhak M. (2023)
Sparsity-Aware Block Diagonal Representation for Subspace Clustering.
31st European Signal Processing Conference (EUSIPCO 2023). Helsinki, Finland (04.-08.09.2023)
doi: 10.23919/EUSIPCO58844.2023.10289969
Konferenzveröffentlichung, Bibliographie
Kurzbeschreibung (Abstract)
A block diagonally structured affinity matrix is an informative prior for subspace clustering which embeds the data points in a union of low-dimensional subspaces. Structuring a block diagonal matrix can be challenging due to the determination of an appropriate sparsity level, especially when outliers and heavy-tailed noise obscure the underlying subspaces. We propose a new sparsity-aware block diagonal representation (SABDR) method that robustly estimates the appropriate sparsity level by leveraging upon the geometrical analysis of the low-dimensional structure in spectral clustering. Specifically, we derive the Euclidean distance between the embeddings of different clusters to develop a computationally efficient density-based clustering algorithm. In this way, the sparsity parameter selection problem is re-formulated as a robust approximation of target between-clusters distances. Comprehensive experiments using real-world data demonstrate the effectiveness of SABDR in different subspace clustering applications.
Typ des Eintrags: | Konferenzveröffentlichung |
---|---|
Erschienen: | 2023 |
Autor(en): | Taştan, Aylin ; Muma, Michael ; Ollila, Esa ; Zoubir, Abdelhak M. |
Art des Eintrags: | Bibliographie |
Titel: | Sparsity-Aware Block Diagonal Representation for Subspace Clustering |
Sprache: | Englisch |
Publikationsjahr: | 1 November 2023 |
Verlag: | IEEE |
Buchtitel: | 31st European Signal Processing Conference (EUSIPCO 2024): Proceedings |
Veranstaltungstitel: | 31st European Signal Processing Conference (EUSIPCO 2023) |
Veranstaltungsort: | Helsinki, Finland |
Veranstaltungsdatum: | 04.-08.09.2023 |
DOI: | 10.23919/EUSIPCO58844.2023.10289969 |
Kurzbeschreibung (Abstract): | A block diagonally structured affinity matrix is an informative prior for subspace clustering which embeds the data points in a union of low-dimensional subspaces. Structuring a block diagonal matrix can be challenging due to the determination of an appropriate sparsity level, especially when outliers and heavy-tailed noise obscure the underlying subspaces. We propose a new sparsity-aware block diagonal representation (SABDR) method that robustly estimates the appropriate sparsity level by leveraging upon the geometrical analysis of the low-dimensional structure in spectral clustering. Specifically, we derive the Euclidean distance between the embeddings of different clusters to develop a computationally efficient density-based clustering algorithm. In this way, the sparsity parameter selection problem is re-formulated as a robust approximation of target between-clusters distances. Comprehensive experiments using real-world data demonstrate the effectiveness of SABDR in different subspace clustering applications. |
Freie Schlagworte: | emergenCITY_CPS |
Fachbereich(e)/-gebiet(e): | 18 Fachbereich Elektrotechnik und Informationstechnik 18 Fachbereich Elektrotechnik und Informationstechnik > Institut für Nachrichtentechnik 18 Fachbereich Elektrotechnik und Informationstechnik > Institut für Nachrichtentechnik > Robust Data Science 18 Fachbereich Elektrotechnik und Informationstechnik > Institut für Nachrichtentechnik > Signalverarbeitung LOEWE LOEWE > LOEWE-Zentren LOEWE > LOEWE-Zentren > emergenCITY |
Hinterlegungsdatum: | 15 Nov 2023 09:20 |
Letzte Änderung: | 25 Jan 2024 06:57 |
PPN: | 514933151 |
Export: | |
Suche nach Titel in: | TUfind oder in Google |
![]() |
Frage zum Eintrag |
Optionen (nur für Redakteure)
![]() |
Redaktionelle Details anzeigen |