[Colloquium] ***Date Correction

Dawn Ellis dellis at ttic.edu
Mon Mar 3 08:59:24 CST 2014


When:     Monday, March 10th at 11am

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

Speaker:  Dan Goldwasser, University of Maryland

Title:       Predicting Real World Outcomes over Structured Latent
Representations

Abstract:

Natural language is a remarkably effective communication tool when used by
humans. However, when used to support interactions with machines, to help
us accomplish useful tasks such as giving instructions to automated agents
or understanding social media discourse, serious challenges arise.
 Approaching these challenges using supervised machine learning methods is
often too costly as the learning tasks become increasingly complex.

In this talk I will present a response-driven approach for natural language
learning problems. Building on the connection between the natural language
input and its use as a real world communication tool, this approach
leverages the real-world result of language interpretation as a supervision
source, thus helping to minimize human involvement and annotation effort.
 From a machine learning perspective, the key challenge is to learn natural
language interpretations without direct supervision at that level. As a
result, the learning process is defined over a set of latent variables
capturing the hidden connections between the natural language input and the
observed real world output.  I will describe the general learning protocol,
and its application for several natural language processing tasks, such as
semantic interpretation and dialog processing.

Bio:

Dan Goldwasser is a postdoctoral researcher working with Hal Daumé, at the
University of Maryland, College Park.  He finished his PhD in 2012, at the
University of Illinois at Urbana Champaign where he was advised by Dan
Roth. Dan's research interest fall in the intersection of natural language
processing and applied machine leaning, focusing on natural language
semantics and discourse processing, and from a machine learning
perspective, structured learning with limited supervision.

Host:  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/20140303/9ad8076c/attachment.htm 


More information about the Colloquium mailing list