Time:2:15-3:30pm, Friday, Sept. 7
Place: Science Center. Room 4102, CUNY Graduate Center. 5th ave
between 34th and 35th Sts.

Title: Structured Perceptron with Inexact Search

Speaker: Liang Huang
Affiliation: Queens College and Graduate Center, City University of New York

Structured learning with inexact inference is a fundamental problem in
machine learning with wide applications in natural language processing
and other structured domains where exact inference is often
intractable, for example in parsing and machine translation.

This work develops a general theory of structured perceptron learning
under inexact inference. We aim to train a search-specific,
search-error-robust model that can "live with" search errors and reach
the true output regardless of how inaccurate the search is.
Specifically, we propose variants of the structured perceptron
algorithm under a general ``violation-fixing'' framework that
guarantees convergence (under new definitions of separability). This
framework subsumes previous remedies including ``early update'' of
Collins and Roark (2004) as special cases (and thus establishing a
theoretical justification for early update), and also explains why
standard perceptron may fail with inexact search. We also propose new
update methods within this framework which learn better models with
dramatically reduced training times on state-of-the-art part-of-speech
tagging and incremental parsing systems.


Liang Huang has just joined CUNY as an Assistant Professor of Computer
Science at Queens College and CS doctoral faculty member at the
Graduate Center. He received his Ph.D. from the University of
Pennsylvania in 2008, and worked briefly at Google as a Research
Scientist before switching to the University of Southern California as
a Research Assistant Professor. His research focuses on efficient
search algorithms for natural language processing, esp. in parsing and
machine translation, as well as related structured learning problems.
His work received a Best Paper Award at ACL 2008, and three Best Paper
Nominations at ACL 2007, EMNLP 2008, and ACL 2010. He is also a
recipient of the Google Faculty Research Award (2010). Besides
research he has a great passion for teaching and received the
prestigious University Prize for Graduate Teaching at Penn.

He is currently looking for bright students and postdocs in NYC so
drop him a note if you are interested.