Date of Award

7-2016

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Dr. Sanjukta Bhowmick

Abstract

My dissertation focuses on developing scalable algorithms for analyzing large complex networks and evaluating how the results alter with changes to the network. Network analysis has become a ubiquitous and very effective tool in big data analysis, particularly for understanding the mechanisms of complex systems that arise in diverse disciplines such as cybersecurity [83], biology [15], sociology [5], and epidemiology [7]. However, data from real-world systems are inherently noisy because they are influenced by fluctuations in experiments, subjective interpretation of data, and limitation of computing resources. Therefore, the corresponding networks are also approximate. This research addresses these issues of obtaining accurate results from large noisy networks efficiently.

My dissertation has four main components. The first component consists of developing efficient and scalable algorithms for centrality computations that produce reliable results on noisy networks. Two novel contributions I made in this area are the development of a group testing [16] based algorithm for identification of high centrality vertices which is extremely faster than current methods, and an algorithm for computing the betweenness centrality of a specific vertex.

The second component consists of developing quantitative metrics to measure how different noise models affect the analysis results. We implemented a uniform perturbation model based on random addition/ deletion of edges of a network. To quantify the stability of a network we investigated the effect that perturbations have on the top-k ranked vertices and the local structure properties of the top ranked vertices.

The third component consists of developing efficient software for network analysis. I have been part of the development of a software package, ESSENS (Extensible, Scalable Software for Evolving NetworkS) [76], that effectively supports our algorithms on large networks.

The fourth component is a literature review of the various noise models that researchers have applied to networks and the methods they have used to quantify the stability, sensitivity, robustness, and reliability of networks.

These four aspects together will lead to efficient, accurate, and highly scalable algorithms for analyzing noisy networks.

Comments

A DISSERTATION Presented to the Faculty of The Graduate College at the University of Nebraska In Partial Fulfillment of Requirements For the Degree of Doctor of Philosophy. Copyright 2016 Vladimir Ufimtsev.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."

COinS