45th Annual Summer Institute of Applied Statistics
Unfortunately, this event has been canceled due to safety concerns regarding COVID-19.
Bayesian Additive Regression Trees: An Introduction and R Tutorial
Predictive methods based on ensembles of trees are a key part of modern data science/machine learning. Random Forests and boosting are two well known ensemble approaches which commonly use simple trees as their basic learner.
Bayesian Additive Regression Trees (BART, Chipman, George, and McCulloch) is a fully Bayesian implementation of ensemble learning using trees. BART is closest to boosting, but the Bayesian approach provides some distinct advantages. Some useful aspects of BART are :
(i) Markov Chain Monte Carlo (MCMC) provides a stochastic search of the model space.
(ii) Uncertainty is capture by the MCMC variation.
(iii) Multiple BART models may be used in a larger hierarchical model.
(iv) prior: simple default prior and simple ways to inject prior information when available.
This BART tutorial will consist of
(a) Review of basic tree modeling, random forests, and boosting.
(b) Basic overview of BART with tutorial on using the R package BART.
(c) Under the hood: how does BART work?
(d) Recent advances in Bayesian ensemble modeling.
To view past presenters, click here.
For questions regarding SIAS, contact Kimri Mansfield at (801) 422-4506 or email@example.com.