Robust Clustering Using Outlier-Sparsity Regularization

Please enable javascript in your browser.

Rahimi, Yaghoub | 2016

1956 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 48485 (02)
University: Sharif University of Technology
Department: Mathematical Sciences
Advisor(s): Daneshgar, Amir
Abstract:
Although clustering algorithms such as k-means and probabilistic clustering are quite popular and widely used nowadays, their performance are too sensitive to the presence of outliers where Even few outliers can compromise the ability of these algorithms to extract hidden data substructures. In this thesis, after going through the basics of some optimization methods such as BCD, EM, and MM, in Section 2 and a review of relevant clustering methods in Section 3, we explore the results of [Forero, et al., Robust clustering using outlier-sparsity regularization. IEEE Trans. Signal Process. (60), 2012] in Sections 4 and 5 where the outliers are handled by introducing a regularization term in the cost function of the corresponding clustering algorithms. They also use kernelized versions of the robust clustering algorithms to efficiently handle high-dimensional data and identify nonlinearly separable clusters. As the main result of this thesis, by an implementation of these algorithms and benchmarking, we show that all algorithms discussed are highly sensitive to the existence of outliers and are not efficient as far as the time complexity is concerned
Keywords:
Clustering ; Outliers ; Data Clustering ; Robust Clustering