A0848
Title: Bayesian sparse Kronecker product decomposition
Authors: Shao-Hsuan Wang - National Central University (Taiwan) [presenting]
Abstract: The sparse Kronecker product decomposition (SKPD) for tensor data was introduced by a recent study. This method represents the first frequentist framework designed for signal region detection in high-resolution, high-order image regression problems. It demonstrated the strong performance of SKPD in various applications. A Bayesian version of SKPD is introduced, referred to as Bayesian SKPD. From a Bayesian perspective, a three-parameter beta-normal prior family is applied to the parameters of interest. Additionally, tensor regression data is addressed with mixed-type responses using polya-gamma augmentation. This approach allows for credible region detection through direct Gibbs sampling. The theoretical results are presented, and the effectiveness of Bayesian SKPD is demonstrated using real brain imaging data from the OASIS.
