Bayesian Analysis for Dimensionality and Complexity Reduction

Zeyu Ding, Katja Ickstadt, Alexander Munteanu

Abstract: In this contribution, we will present Bayesian approaches for dimensionality and complexity reduction in the context of health-related problems. Following an introduction to Bayesian analysis in general, we will first show two examples of Bayesian variable selection methods for reducing the number of variables, one for binary data with an application to Single Nucleotide Polymorphisms (SNPs) for the HapMap dataset, and one for time-to-event endpoints with an application to glioblastoma data from the Cancer Genome Atlas. Second, we will present an approach for reducing statistical models, where we transfer the Merge & Reduce principle to maintain statistical summaries in streaming models. The variable selection approaches as well as the Merge & Reduce approach are important steps towards resource-aware data analyses.