A0698
Title: On prior distributions for orthogonal function sequence: Application to Bayesian functional principle component analysis
Authors: Shonosuke Sugasawa - Keio University (Japan) [presenting]
Daichi Mochihashi - The Institute of Statiatical Mathematics (Japan)
Abstract: The aim is to propose a novel class of prior distributions for sequences of unknown functions that encourages near-orthogonality via a basis function expansion. Such prior constructions are particularly useful in Bayesian approaches to functional principal component analysis (FPCA), where orthogonality among principal component functions is crucial for ensuring a parsimonious representation of functional variation. The framework allows not only for uncertainty quantification of functional estimates but also for automatic selection of the number of components, providing a fully probabilistic treatment of FPCA. The flexibility and effectiveness of the proposed methodology are demonstrated through numerical experiments and real data applications.
