Secuiu, Andrei-Rafael (2026) Efficiency Improvement of Particle Filtering Algorithms within the Pairwise Sequentially Markovian Coalescent Framework. Master's Thesis / Essay, Applied Mathematics.
|
Text
MScThesisAndreiSecuiu.pdf Download (14MB) | Preview |
Abstract
The study of ancient demographics of human populations provides fundamental insights into our evolutionary past. An approach complementary to archaeology relies on contemporary genetic data and sophisticated mathematical modelling. A popular framework for inference is the Pairwise Sequentially Markovian Coalescent (PSMC). A particular inference method relies on a sequential Monte Carlo sampling technique called Particle Filtering (PF), which explores the space of complex posterior distributions by iteratively constructing weighted samples denoted particles. Algorithmic heuristics suggest that the promotion of sampling diversity may prove beneficial to the computational efficiency of PF within the PSMC framework. In this thesis, a method to promote sampling diversity is constructed, and its effects on the quality of inference across various fixed particles counts are investigated. Results indicate that the effects of diversity, as implemented in the present work, are mixed and generally exacerbate the self-normalized Importance Sampling bias inherent in the baseline PF method. Mechanistic explanations for this failure are hypothesised, and alternative strategies for promoting diversity are proposed.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Grzegorczyk, M.A. and Lunter, G.A. |
| Degree programme: | Applied Mathematics |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 26 Aug 2026 14:38 |
| Last Modified: | 26 Aug 2026 14:38 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/37516 |
Actions (login required)
 |
View Item |