Subject Area
Computer Science
Abstract
We present several Granger-inspired methodologies that broaden the applicability of traditional Granger causality analysis beyond its conventional use in simple cause-and-effect detection. While Granger causality is often framed in causal terms, it is more appropriately understood as a methodology for identifying predictability and temporal structure between variables. In this work, we investigate temporal structure in sequential datasets, including but not limited to classical time-series data. We introduce new Granger-based methods for selecting the lag structure that characterizes predictive relationships, and we demonstrate how these methods can be used to inform Granger-inspired clustering procedures and neural network architectural design. We further explore the use of objective functions augmented with Granger-based regularization terms in the context of time-series representation learning. Finally, we extend the notion of Granger causality as a tool for temporal structure detection to the evaluation of random bitstreams, showing that Granger-inspired tests provide a new mechanism for identifying and quantifying predictability in sequences otherwise assumed to be random.
Degree Date
Spring 5-16-2026
Document Type
Dissertation
Degree Name
Ph.D.
Department
Computer Science
Advisor
Eric C. Larson
Second Advisor
Mitchell A. Thornton
Third Advisor
Chen Wang
Fourth Advisor
Matthew J. Hornbach
Fifth Advisor
Micah A. Thornton
Number of Pages
282
Format
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Recommended Citation
Sylvester, Joshua H., "Beyond Time Series: Extending Granger Causality for Clustering, Representation Learning, and Randomness Testing" (2026). Computer Science and Engineering Theses and Dissertations. 55.
https://scholar.smu.edu/engineering_compsci_etds/55
