Year: 2006
Type: dissertation
Abstract: In classical clustering, each data point is assigned to at least one cluster. However, \nin many real-world problems, only a small subset of the data clusters well, while the \nrest shows little or no clustering tendencies. For such situations, this thesis presents \nseveral techniques that cluster only a subset of the data into one or more groupings. \nWe first develop a very general parametric approach called Bregman Bubble \nClustering that can find multiple dense regions, and can scale to very large datasets. \nBy using a fast iterative relocation based approach combined with a novel concept \nfor improving local search called Pressurization, Bregman Bubble Clustering extends \ndensity-based clustering to a much larger set of problems. We also develop a seeding \nalgorithm that can automatically determine the number of clusters, and make the \nviii \nresults deterministic. \nWe then describe a more focussed non-parametric alternative called Automated \nHierarchical Density Shaving (Auto-HDS), a framework that consists of a \nfast, hierarchical, density-based clustering algorithm and an unsupervised model selection \nstrategy. Auto-HDS can automatically select between clusters of different \ndensities, present them in a compact hierarchy, and rank individual clusters using \nan innovative stability criteria. The Auto-HDS framework also provides a simple yet \npowerful 2-D visualization of the hierarchy of clusters that is useful for further exploring \nthe dense clusters in high-dimensional datasets. We also developed a robust, \nmemory efficient, platform independent, and open source Java based implementation \nof Auto-HDS called Gene DIVER (Gene Density Interactive Visual Explorer) that \nprovides interactive clustering capabilities for high-throughput biological datasets. \nFor problems where finding small dense regions is important, the parametric \napproach is applicable to a wide variety of scenarios and is scalable to very large \ndatasets. On the other hand, Auto-HDS, the non-parametric approach, provides \na powerful visualization, a compact clustering hierarchy, and interactive clustering: \nproperties that are useful for biologists interested in finding and understanding small \ndense clusters of genes. Together, the two approaches greatly extend the scope of \ndensity based clustering in three different dimensions; the diversity of problems that \ndensity-based clustering can now be used with, the expanded capability to quickly \nunderstand and analyze the clusters in the data, and the scale of the problems that \nare now within reach of modest computing resources.
Author: Gunjan Gupta
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W1899843818