Skip to Content

Department of Cognitive Science

Seminar Abstract

Discrete non-parametric methods. (CLaS-CCD Research Colloquium Series 2012)

Speaker : Dr Wray Buntine, Statistical Machine Learning Group, NICTA.
Date : 21st of November 2012, 2:00PM until 3:00PM
Location : E6A 357, Macquarie University.

The talk is motivated by the need to model networks of probability vectors, used throughout general tasks in machine learning and document analysis. Examples will be given in unsupervised part-of-speech induction and topic modelling over a time-series of documents (but not analysed in detail). The standard non-parametric technique for handling this problem uses variations of the Dirichlet Process, the Pitman-Yor process, the Chinese Restaurant process and various hierarchical extensions. This tutorial will (attempt to) explain how to understand and work with these techniques. We will consider how these various processes work, how one models with them, and how to develop algorithms using them. The aim is to do this is a more simplified manner than is usually done in the literature.