Fast Immune Greedy Spectral Clustering

The performance of the classical spectral clustering algorithm has bottleneck problem of requiring a great deal computation memory and computation time, which makes spectral clustering method limited in large-scale dataset application. To solve this, a novel algorithm called Immune Greedy Spectral C...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:International Information Institute (Tokyo). Information Ročník 15; číslo 1; s. 375
Hlavní autori: Gou, S P, Zhang, J, Jiao, L C, Yang, J Y
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Koganei International Information Institute 01.01.2012
Predmet:
Algorithms
Approximation
Clustering
Datasets
Eigenvectors
ISSN:1343-4500, 1344-8994
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:The performance of the classical spectral clustering algorithm has bottleneck problem of requiring a great deal computation memory and computation time, which makes spectral clustering method limited in large-scale dataset application. To solve this, a novel algorithm called Immune Greedy Spectral Clustering (IGSC) algorithm is proposed, in which, the clonal selection algorithm is introduced to select the typical sample sets without any prior knowledge and to decrease the computational complexity of similarity matrix for eigenvector. The proposed algorithm substitutes a subset for all datasets to eigenvector decompose, which is regard as a combinatorial optimization issue in immune clonal selection algorithm. And immune clonal selection algorithm ensures a fast convergence to the global optimal solution so that the IGSC algorithm can achieve more stable and better clustering result The experiment results on UCI datasets, texture images show that IGSC algorithm needs no prior knowledge and saves much more time compared with Greedy Spectral Clustering (GSC) algorithm while also gets a better accuracy rate compared with the standard Nyström Spectral Clustering (NSC) algorithm. [PUBLICATION ABSTRACT]
Bibliografia:SourceType-Scholarly Journals-1
ObjectType-Feature-1
content type line 14
ISSN:1343-4500
1344-8994