[Colloquium] TTIC Colloquium: Lyle Ungar, University of Pennsylvania

[Dawn Ellis]

When:     Monday, March 2nd at 11am

Where:    TTIC, 6045 S Kenwood Avenue, 5th Floor, Room 526

Who:       Lyle Ungar, University of Pennsylvania

Title:       Spectral Language Modeling


Low dimensional vector models of language have long been built using
Singular Value Decomposition (SVD) of word co-occurrence data.  More
recently, such methods have been extended to build optimal linear
estimators of hidden state in probabilistic models such as HMMs and
PCFGs. These spectral methods are much faster than EM and come with
optimality proofs.  We present a fast, clean spectral method for
estimating HMMs, sketch some generalizations of it, and show that these
spectral methods can be used to improve performance in a wide range
natural language processing (NLP) applications.


joint work with Paramveer Dhillon, Jordan Rodu, Mike Collins, and Dean
Foster

Hosts:  David McAllester, mcallester at ttic.edu  and  Karen Livescu,
klivescu at ttic.edu

-- 
*Dawn Ellis*
Administrative Coordinator,
Bookkeeper
773-834-1757
dellis at ttic.edu

TTIC
6045 S. Kenwood Ave.
Chicago, IL. 60637
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20150223/ad64963d/attachment.htm