National Technical University of Athens (NTUA)
Δυναμικά μεταγενέστερες πρότερες κατανομές σε προβλήματα συρρίκνωσης
Abstract
dc:descriptionIn this PhD thesis, we focus on the problem of Bayesian variable selection in high-dimensional linear regression models. In cases where the sample size n is smaller than the number of explanatory variables p, the use of shrinkage priors is a well-defined methodology with eminent theoretical and empirical properties. With the use of imaginary samples and shrinkage priors as baseline priors under the Power-Expected-Posterior (PEP) prior methodology, we construct compatible shrinkage priors with appealing interpretation. Furthermore, we examine the idea of augmenting the imaginary design matrix to make its columns orthogonal, in order to create independent PEP shrinkage priors based on predefined, non-subjective initial priors. Under this methodology, we employ appropriate prior distributions on the power hyperparameters of the PEP prior methodology, leading to mixtures of independent prior distributions suitable for variable selection problems when n << p. This second approach offers algorithmic flexibility and low computational cost. We investigate the theoretical properties of the proposed methodologies as well as their performance on both simulated and real-world datasets.
Degree
thesis:*- Grantor dc:publisher
- National Technical University of Athens (NTUA)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tzoumerkas, Georgios
- Τζουμέρκας, Γεώργιος
Subjects
dc:subject × 16- Μπεϋζιανή επιλογή μεταβλητών
- Δυναμικά-Μεταγενέστερες-Αναμενόμενες εκ των προτέρων κατανομές
- ΡΕΡ-πρότερες συρρίκνωσης
- Αραιά δεδομένα
- Γραμμική παλινδρόμηση μεγάλων διαστάσεων
- Bayesian variable selection
- Power-Expected-Posterior (PEP) priors
- PEP-Shrinkage priors
- Sparse datasets
- High-dimensional linear regression
- Μαθηματικά
- Φυσικές Επιστήμες
- Στατιστική και Πιθανότητες
- Mathematics
- Natural Sciences
- Statistics and Probability
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58191