[Compstats] Talk Wed *4:30pm* in MC6007: Bayesian Language Modelling

Pascal Poupart ppoupart at cs.uwaterloo.ca
Tue Jan 26 18:40:17 EST 2010


Since the AI seminar room is not free, we will hold the seminar in 
MC6007 at 4:30pm.  Please note also the time change: *4:30pm* (instead 
of 4pm).


Pascal Poupart wrote:
> Hi,
> As promised, I will give the second compstats seminar of the year this 
> Wednesday @ 4pm.  I will talk about Bayesian methods for language 
> modelling and some research in collaboration with Google to extract 
> short synthesizing phrases to label clusters of documents in an 
> unsupervised fashion.  Title and abstract follow.
> cheers,
> Pascal
> ============================
> Title: Bayesian Language Modelling
> Speaker: Pascal Poupart
> Date: Wednesday, Jan 27
> Time: 4:30pm
> Location: MC6007
> Recent years have seen an increased use of Bayesian techniques for 
> language modelling due to their flexibility, their ability to model 
> various syntactic and semantic sructure and the possibility to include 
> prior knowledge.  In this talk, I will first review existing 
> techniques for topic modelling such as probabilistic latent semantic 
> analysis, latent Dirichlet allocation and hierarchical Dirichlet 
> processes.  I will also review co-location techniques such as n-gram 
> models and the hierarchical Pitman-Yor process.  I will then discuss 
> how topic models and co-location models can be combined to yield 
> powerful language models that simultaneously capture some elements of 
> syntax and semantics.  I will show the benefits of such models to 
> extract short synthesizing phrases to label clusters of documents in 
> an unsupervised fashion.
> If time permits, I will also discuss some of the challenges to extend 
> these models to deal with entities and relations that are often 
> thoughts as the pillars of deep natural language understanding.  I may 
> also discuss some of the challenges to learn these models in an online 
> fashion by sequential monte carlo techniques.
> This work was done in collaboration with Ting Liu (UW), Ruitong Huang 
> (UW), Andy Chiu (Google) and Finnegan Southey (Google)
> This work is supported financially by a Google Research Award

Pascal Poupart
Associate Professor
David R. Cheriton School of Computer Science
University of Waterloo
200 University Avenue West
Waterloo, Ontario
Canada N2L 3G1
Web: http://www.cs.uwaterloo.ca/~ppoupart
Email: ppoupart at cs.uwaterloo.ca 
Telephone: 1-519-888-4567x36239 
Fax: 1-519-885-1208

More information about the Compstats mailing list