About this Class

Nonparametric Bayesian methods are a class of probabilistic models that automatically adjust their complexity based on the data they are used to model.  As a simple example, the Dirichlet Process (a particular nonparametic Bayesian prior that will be our launching point) can be used to perform clustering when the number of clusters is not known a priori.  This seminar is a deep exploration of models of this type, from high level reasoning down to concrete implementation, with examples drawn from computational linguistics, computational psycholinguistics and computer vision.

The course should be accessible to students who have taken any of: Computational Linguistics I, Computational Psycholinguistics, or any rigorous course in probability/statistics or machine learning.  Please talk to one of the instructors if you are in doubt.  This course will involve a substantial amount of reading, as well as large course projects of your choosing.