The Department of Statistics at Iowa State University hosted a seminar featuring Huiyan Sang from Texas A&M University. Sang presented the GS-BART method, focusing on its application in spatial and network data analysis. She emphasized the necessity for improved methodologies to handle complex data with graph relations.
GS-BART, or Graph Split Additive Decision Trees, offers a novel approach to enhance Bayesian additive decision trees. By incorporating a flexible split rule aligned with graph structure, it relaxes assumptions common in existing ensemble decision tree models. Sang's team designed a scalable informed MCMC algorithm, enabling efficient sampling of the graph-split-based decision tree.
During the seminar, Sang showcased GS-BART's performance compared to traditional ensemble tree models and Gaussian process models. The method demonstrated efficacy across various regression and classification tasks tailored for spatial and network data analysis. Sang, a distinguished professor at Texas A&M University, has extensive expertise in statistics, with interdisciplinary research spanning environmental sciences, geosciences, economics, and biomedical research.
Attendees gained valuable insights into cutting-edge statistical methodologies, witnessing the potential of GS-BART to advance data analysis in spatial and network contexts. Sang's presentation underscored the importance of innovative approaches to tackle the complexities of modern datasets, offering a glimpse into the future of statistical research and application.