Events

At the heart of Bayesian statistics lies the concept of conditional probability. The notion of a posterior distribution enables us to probabilistically incorporate our prior beliefs about unknown quantities with informative data. This is fundamental in Bayesian machine learning, facilitating applications in topic models, recommendation systems, and natural language processing.
The Dirichlet multinomial framework serves as a classic and important example of a flexible Bayesian model. Bayesian nonparametric statistics elevates this from random vectors to the calculus of random measures, with the Dirichlet process (DP) as a seminal contribution. My work has focused on advancing modelling capabilities beyond the DP, enhancing practical applications and formal analyses.
In this talk, I will explore the origins of the Poisson Partition Calculus, a method I developed for analyzing a wide range of processes, beginning with a 2002 arXiv preprint and culminating in a series of papers. I will also highlight my highly influential collaborations with Hemant Ishwaran (2001 JASA) on topics we named and popularized: stick-breaking priors and Pitman-Yor processes, which have roots in more esoteric excursion theory.
Underlying this discussion is how I have been influenced by great and generous people throughout my career, starting with my advisor Albert Lo.

****