Files and links (1)
Thesis_Alan_Arakkal2.73 MB
Embargoed Access, Embargo ends: 01/23/2027
Dissertation
JANE: Just Another latent space NEtwork clustering algorithm
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Autumn 2025
DOI: 10.25820/etd.008194
Abstract
While latent space network models have been a popular approach for community detection for over 15 years, major computational and methodological challenges remain. This dissertation introduces novel methodologies to address these limitations, including a fast generalized expectation-maximization algorithm that incorporates a low-dimensional approximation to adjust for degree heterogeneity, an approximation of intractable likelihood terms via Taylor expansions, a fast graphical neural network initialization algorithm, and convergence criteria focused on clustering performance. Additionally, we develop the Latent Space Hurdle Model, a generalization of the standard latent space model, to account for noise edges by jointly modeling the latent propensity to form an edge and the observed edge weight, enabling probabilistic down-weighting of spurious connections and improving clustering accuracy. Simulation studies demonstrate substantial gains in both computational efficiency and clustering performance. Finally, using an agent-based model, we apply the methods developed in this dissertation to evaluate community detection as a quarantining protocol for mitigating disease spread, demonstrating the effectiveness of targeting densely connected clusters to disrupt transmission pathways.
Details
- Title: Subtitle
- JANE: Just Another latent space NEtwork clustering algorithm
- Creators
- Alan Tom Arakkal
- Contributors
- Daniel K Sewell (Advisor)Brian J Smith (Committee Member)Grant D Brown (Committee Member)Joseph E Cavanaugh (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Biostatistics
- Date degree season
- Autumn 2025
- DOI
- 10.25820/etd.008194
- Publisher
- University of Iowa
- Number of pages
- x, 142 pages
- Copyright
- Copyright 2025 Alan Tom Arakkal
- Comment
- This thesis has been optimized for improved web viewing. If you require the original version, contact the University Archives at the University of Iowa: https://www.lib.uiowa.edu/sc/contact/
- Language
- English
- Date submitted
- 11/21/2025
- Description illustrations
- illustrations, tables, graphs
- Description bibliographic
- Includes bibliographical references (pages 139-142).
- Public Abstract (ETD)
- Network analysis examines the relationships or connections between entities (i.e., actors) and the patterns that emerge from their interactions. A key aspect of network analysis is community detection, which identifies groups of closely connected actors within networks and is widely used in fields ranging from social science to public health. Traditional approaches for detecting communities using latent space models are often slow and can be distorted by spurious or noisy connections. This dissertation introduces new methods to address these challenges, including a fast algorithm for fitting latent space models and an extension called the Latent Space Hurdle Model, which improves accuracy by accounting for noisy connections. Simulation studies show that these methods are both faster and more accurate than existing approaches. Finally, using an agent-based model, we demonstrate the utility of community detection to guide quarantining strategies during disease outbreaks, showing that targeting densely connected clusters can effectively reduce disease transmission.
- Academic Unit
- Biostatistics
- Record Identifier
- 9985134944902771
Metrics
1 Record Views