Discrete Non-Parametric Models for Natural Language Processing

Wray Buntine (NICTA)

NICTA SML SEMINAR Machine Learning Research Group Seminar

DATE: 2013-03-07
TIME: 11:15:00 - 12:00:00
LOCATION: CRL - NICTA Seminar Room, Ground Floor
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
Recent successes in core natural language processing tasks have seen a wave of successes using Bayesian non-parametric methods: document segmentation, unsupervised part of speech induction, unsupervised key-phrase recognition, etc.

This talk will give a general, non-technical tour of some of the models and methods behind these methods focusing on the Pitman-Yor (or Poisson-Dirichlet) Process. Probabilistic graphical models will be used to present some models and some modelling and development issues will be presented.

BIO:

Updated: 5 March 2013 / Responsible Officer: JavaScript must be enabled to display this email address. / Page Contact: JavaScript must be enabled to display this email address. / Powered by: Snorkel 1.4