BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Nonparametric Bayesian Natural Language Model Domain Adaptation: A
  Hierarchical\, Hierarchical Pitman-Yor Process Language Model - Dr Frank 
 Wood (UCL)
DTSTART:20080917T130000Z
DTEND:20080917T140000Z
UID:TALK12903@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:There are many real-world modeling problems for which one may 
 not have\na sufficient quantity of training data to reliably estimate a us
 eful\nmodel. Obtaining sufficient quantities of training data for these\n"
 specific" modeling domains can be both costly as well as a\nsignificant lo
 gistical challenge. In some cases there may already\nexist a large quantit
 y of training data from a related or more general\ndomain. The phrase doma
 in adaptation is used to describe modeling\ntechniques that utilize such d
 ata (copious but general) to improve\nmodeling of specific domains for whi
 ch training data is not as readily\navailable.  Various language model dom
 ain adaptation approaches have\nbeen proposed\; however\, this work is the
  first to show how to do\ndomain adaptation of hierarchical nonparametric 
 Bayesian language\nmodels.\n\nSpecifically we define a hierarchy of hierar
 chical Pitman-Yor process\nlanguage model and explain how such a model acc
 omplishes domain\nadaptation (intuitively\, it "backs-off" to both in- and
  out-of-domain\nmodels in a way that is similar in spirit to the backing-o
 ff that\nsmoothed n-gram models do within a single domain). For estimation
  and\ninference we define a novel multi-floor Chinese restaurant franchise
 \nrepresentation and sampler.  Encouragingly\, for various natural\nlangua
 ge corpora we find that our new approach to domain adaptation\noutperforms
  all of the existing approaches against which it was\ncompared.
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR