Networks are everywhere, from the financial systems powering global economies to the social connections that shape communities and the infrastructure that moves people and goods. This dissertation examines the integration of network science, operations research, and machine learning to develop methodological contributions that advance the modeling, optimization, and interpretation of complex systems. These contributions are presented across four chapters: • Chapter 2 introduces the Feature-Based Network Construction (FBNC) framework for reconstructing networks from partial or aggregate data, enabling exact feature-constrained sampling and providing new tools for both analysis and “what-if” scenario exploration. • Chapter 3 investigates network connectivity optimization through the lens of Markov chains, proposing an algorithm that directly minimizes mean first passage times while remaining robust to uncertainty in edge presence. • Chapter 4 presents CoNNect, a connectivity-preserving regularization method that enforces computational efficiency through sparsity in neural networks without sacrificing expressiveness or performance. • Chapter 5 develops a clustering framework to identify the functional positions of financial institutions within multi-layer financial networks, offering regulators interpretable insights into systemic roles such as intermediaries, connectors, and peripheral actors.
More information on the thesis