A tale of two (thousand) trees in nonparametric modeling

Abstract: In this talk I will explore the roles recursive dyadic trees play in two classes of Bayesian nonparametric models---(i) tree-structured random measures such as the Polya tree and its extensions and (ii) discrete mixture models based on stick-breaking processes. I will discuss similarities and distinctions in the ways the underlying dyadic tree impacts the resulting inference in these two contexts. Then based on these considerations, I will present some recent work on scaling up tree-structured nonparametric models to handle high-dimensional data in unsupervised learning.

Zoom link: LINK