Public defence in Computer Science, M.Sc. Onur Poyraz

Title of the thesis: Flexible Bayesian latent variable modeling of interacting processes in healthcare time-series

Thesis defender: Onur Poyraz
Opponent: Docent, University Lecturer Moritz Schauer, Chalmers University of Technology, Sweden
Custos: Associate Professor Pekka Marttinen, Aalto University School of Science, Department of Computer Science

Modern healthcare is generating more data than ever before. From daily blood sugar measurements to the patterns of infections across the body, these continuous information streams hold valuable insights into human health. Yet, making sense of how different biological processes interact over time remains one of the major challenges in medicine and public health today. Understanding these complex relationships is essential for advancing personalized care and predicting health outcomes more accurately.

The motivation behind this thesis stems from the growing realization that traditional models often struggle to capture the intricate ways in which biological systems influence one another. In real life, health is rarely the result of a single factor; instead, it emerges from the constant interplay of many processes. Existing tools are often too rigid, simplistic, or slow to keep pace with the complexity and volume of modern healthcare data. There is a pressing need for more flexible, scalable, and interpretable methods that can reveal the hidden dynamics governing patient health, both at the individual and population levels.

This thesis responds to this need by developing a suite of new computational methods explicitly designed to model interacting processes. The contributions include innovations that make it possible to represent more complex dependencies between biological systems, improvements that allow models to work efficiently even with large-scale data, and frameworks that capture how individuals can differ from overall population trends. These methods have been validated using real-world clinical datasets, demonstrating significant advances in both accuracy and scalability. The results offer powerful new tools for healthcare researchers, clinicians, and public health experts. By helping to uncover and understand the invisible threads connecting different aspects of human health, this research brings the promise of more personalized, predictive, and effective healthcare a step closer to reality.

Probabilistic graphical models, Markov chain Monte Carlo, Coupled hidden Markov models, Gaussian processes, Differential equations, Interacting processes, Healthcare time-series

Thesis available for public display 10 days prior to the defence at Aaltodoc. 

Contact information of doctoral student: onur.poyraz@aalto.fi