Graphpcor: Prior for Correlation Matrices
This talk introduces a scalable, graph-based framework for modeling correlation matrices that integrate expert-informed priors.
Overview
The correlation coefficient is an important summary of the dependency between two variables. When studying several variables, the correlation matrix summarizes the marginal linear relationships among them. In data learning a correlation matrix some challenges arise, including rapid growth in the number of parameters. We address this challenge by proposing an approach to model correlation matrices using a graph to represent conditional dependency assumptions. We show that our approach requires as many parameters as the number of edges in the graph. We further leverage expert knowledge by introducing a prior that penalizes divergence from a base correlation matrix. When combined, it provides an informative model-based prior for correlation matrices. We will illustrate our approach with an application.
Presenters
Brief Biography
Elias Teixeira Krainski is a Research Scientist working in the Bayesian Computation group. He received his Ph.D. from the Norwegian University of Science and Technology in Trondheim, Norway. He previously lectured as Adjunct Professor in the Statistics department of the Federal University of Paraná, in Curitiba, Brazil.
