Title: | LVCluster: Bounded Clustering using Laguerre Voronoi Diagram |
Authors: | Ohi, Abu Quwsar Gavrilova, Marina |
Citation: | WSCG 2024: full papers proceedings: 32. International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, p. 255-262. |
Issue Date: | 2024 |
Publisher: | Václav Skala - UNION Agency |
Document type: | konferenční příspěvek conferenceObject |
URI: | http://hdl.handle.net/11025/57397 |
ISSN: | 2464–4625 (online) 2464–4617 (print) |
Keywords: | Laguerrova geometrie;Voronoiův diagram;shlukování;KMeans;klesající gradient |
Keywords in different language: | Laguerre Geometry;Voronoi Diagram;clustering;KMeans;gradient descending |
Abstract in different language: | Clustering, a fundamental technique in unsupervised learning, identifies similar groups within a dataset. However, clustering algorithms encounter limitations when requiring a predetermined number of clusters/centroids/labels. This paper proposes a novel approach of clustering by integrating concepts from Voronoi diagrams in Laguerre geometry, namely, Laguerre Voronoi Clustering (LVCluster). Laguerre geometry introduces circles by adding radius weight metric to centroids, enabling dynamic exclusion from clustering criteria. Consequently, this approach offers flexibility by necessitating only one hyperparameter, an upper-bound value for the number of circles. LVCluster can be optimized using gradient descent and can be jointly optimized with deep neural network architectures. The experimental results indicated that LVCluster outperforms clustering algorithms when trained individually and jointly with deep neural networks on increased cluster centroids. |
Rights: | © Václav Skala - UNION Agency |
Appears in Collections: | WSCG 2024: Full Papers Proceedings |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
C53-2024.pdf | Plný text | 1,21 MB | Adobe PDF | View/Open |
Please use this identifier to cite or link to this item:
http://hdl.handle.net/11025/57397
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.