Files in this item



application/pdfTHOMPSON-THESIS-2019.pdf (2MB)
(no description provided)PDF


Title:Data-centric methods for optimization and pattern discovery in networked systems
Author(s):Thompson, Karl Hany
Advisor(s):Tran, Huy T
Department / Program:Aerospace Engineering
Discipline:Aerospace Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Machine Learning
Operational Research
Unsupervised Learning
Air Network
Recurrent Networks
Abstract:In this thesis, we examine two data-driven solutions to problems in operational networks. The first problem is concerned with assessing the resilience of the US air transportation network from an operational perspective. As a complex network comprising over 5,000 public airports and countless interfaces with other transportation systems, the impact of any disruption in the air network undoubtedly extends to other inter-connected economic and functional domains. Our solution to the resilience assessment problem is a tri-level optimization program that is able to simulate worst-case disruptions in the air network as well as propose the optimal ways to mitigate their effects. These mitigation steps take the form of investment recommendations for the air routes that are in most need of augmentation by other high-speed transportation modes. Our methodology and results for this application are explained in detail in Chapter 3. The second problem discussed in this thesis is centered on identifying design patterns in architecture graph representations of operational systems. Design patterns have been well documented and researched in software systems as a valuable design tool since the nineties. However, their usage has not been significantly expanded beyond software architectures, and their discovery methods have generally remained structured and supervised. We propose an end-to-end, unsupervised graph generation and pattern identification framework that is able to find unknown and potentially useful patterns in architecture graphs using machine learning. Our method is not limited to software systems, and is designed to be able to make possible pattern predictions even with a single architecture graph input. We detail our framework and experimental results in Chapter 4. Organizationally, Chapter 1 of the thesis starts with an introduction to network theory and graph representations, Chapter 2 provides background on network optimization and graph machine learning, and Chapter 5 concludes the thesis with our final thoughts and future research directions.
Issue Date:2019-07-19
Rights Information:Copyright 2019 Karl H. Thompson
Date Available in IDEALS:2019-11-26
Date Deposited:2019-08

This item appears in the following Collection(s)

Item Statistics