Vector quantization based approximate spectral clustering of large datasets

Spectral partitioning, recently popular for unsupervised clustering, is infeasible for large datasets due to its computational complexity and memory requirement. Therefore, approximate spectral clustering of data representatives (selected by various sampling methods) was used. Alternatively, we propose to use neural networks (self-organizing maps and neural gas), which are shown successful in quantization with minimum distortion, as preliminary sampling for approximate spectral clustering. We show that they often outperform other sampling methods, in terms of clustering accuracy. Additionally, we introduce a local density-based similarity measure—constructed without user-set parameters—which achieves accuracies superior to the accuracies of commonly used distance based similarity

TASDEMIR Kadim; 

2012-04-10

ELSEVIER SCI LTD

JRC68885

0031-3203,   

http://www.sciencedirect.com/science/article/pii/S003132031200074X,    https://publications.jrc.ec.europa.eu/repository/handle/JRC68885,   

10.1016/j.patcog.2012.02.012,