{"success":1,"msg":"","color":"rgb(28, 35, 49)","title":"On Theory and Practice for Bayesian Tree Ensembles<\/b>","description":"webinar","title2":"","start":"2021-03-15 04:00","end":"2021-03-15 05:00","responsable":"Dan Kowal <\/i><\/a>","speaker":"Antonio Linero","id":"31","type":"webinar","timezone":"America\/Chicago","activity":"Register in advance for this meeting: https:\/\/riceuniversity.zoom.us\/meeting\/register\/tJMsfuutrzoqGtY4dtclB-TSPWnetfo54YTV","abstract":"Bayesian decision tree ensembles have seen a surge in popularity in recent years, particularly due to their strong performance at uncertainty quantification in causal inference problems. In this talk, we survey recent theoretical and practical advances for this class of models. Applications include high-dimensional variable selection, conditional distribution estimation, and survival analysis. A desirable feature of the models we present is that they allow for analysts to smoothly interpolate between parametric and nonparametric models, allowing the data to determine an appropriate level of complexity for a given problem. We also show that Bayesian additive regression trees possess very strong theoretical properties, obtaining near-minimax optimal rates of convergence adaptively over a large class of function spaces. Our results give a theoretical justification to a heuristic interpretation for the effectiveness of tree ensembles: they induce shrinkage towards functions with low-order interactions."}