Title: Bayesian analysis of dependent functional data
Authors: Silvia Montagna - University of Turin (Italy) [presenting]
Surya Tokdar - Duke University (United States)
Irina Irincheeva - University of Bern (Switzerland)
Abstract: The rapid evolution of data collection technologies has permitted vast quantities of data to be recorded densely over time or space. These data are usually regarded as error-prone measurements of underlying smooth functions, thus the name of functional data. A new Bayesian methodology is proposed for dependent functional data analysis. Dependency manifests across multiple trajectories at any given time point and, potentially, also temporally within each trajectory (e.g. gene expression trajectories). To accommodate for the dependence across curves, we will focus on hierarchical Bayesian factor models. Given the high dimensionality of the data, it will also be necessary to ensure parsimony and computational tractability of the proposed methods. We will explore the use of sparsity-inducing priors, such as latent threshold priors, for certain model parameters. The proposed methods will be implemented on real data applications.