A0763
Title: Dependent beta process in Bayesian factor analysis
Authors: Junyi Zhang - Bocconi University (Italy) [presenting]
Abstract: Bayesian factor models are a popular tool for factor analysis. Current state-of-the-art Bayesian factor analysis approaches leverage the beta-Bernoulli process prior to characterize the factors. This prior, however, ignores the diversity across multiple populations. A new framework based on the dependent beta processes is presented, which allows the development of an innovative hierarchical modeling methodology for Bayesian factor analysis. The distributional properties of the dependent beta process are discussed. Then, the dependent beta process fits into the celebrated beta process factor analysis model and devises the MCMC posterior inference schemes. The usefulness of the hierarchical modeling methodology is illustrated using a psychological assessment dataset.
