Thursday, January 21, 2021 - 14:00 to 16:00
SCHOOL OF COMPUTER SCIENCE
The School of Computer Science is pleased to present…
MSc Thesis Defense by: Nachiket Bhide
Date: Thursday January 21, 2021
Time: 2:00pm – 4:00pm
Zoom URL: https://zoom.us/j/94538079480?
Passcode: If interested in attending this event, contact the Graduate Secretary at csgradinfo@uwindsor.ca
Abstract:
Several techniques are used for clustering of high-dimensional data. Traditionally, clustering approaches are based on performing dimensionality reduction of high-dimensional data followed by classical clustering such as $k$-means in lower dimensions. However, this approach based on $k$-means does not guarantee optimality. Moreover, the result of $k$-means is highly dependent on initialization of cluster centers and hence not repeatable, while not being optimal. To overcome this drawback, an optimal clustering approach in one dimension based on dimensionality reduction is proposed. The one-dimensional representation of high dimensional data is obtained using Kernel Principal Component Analysis. The one-dimensional representation of the data is then clustered optimally using a dynamic programming algorithm in polynomial time. Clusters in the one-dimensional data are obtained by minimizing the sum of within-class variance while maximizing the sum of between-class variance. The advantage of the proposed approach is demonstrated on synthetic and real-life datasets over standard $k$-means in terms of optimality and repeatability.
Keywords: Dimensionality Reduction, Kernel Principal Component Analysis, Optimal Clustering, Dynamic Programming, k-Means
MSc Thesis Committee:
Internal Reader: Dr. Alioune Ngom
External Reader: Dr. Dilian Yang
Advisor: Dr. Luis Rueda
Chair: Dr. Sherif Saad Ahmed
MSc Thesis Defense Announcement
5113 Lambton Tower 401 Sunset Ave. Windsor ON, N9B 3P4 (519) 253-3000 Ext. 3716 csgradinfo@uwindsor.ca