AIC Seminar Series
Amortized inference for structured output prediction
Notice: Hosted by Rodrigo Braz
Date: 2013-11-07 at 16:30
Location: EK255 (SRI E building) (Directions)
Structured output prediction in NLP consists of using a learned model
with an inference algorithm to predict structures for inputs.
Typically, at prediction time, each example is considered to be
independent of all other examples.
In this talk, I will argue that we can take advantage of regularities
of outputs over entire datasets to give an amortized improvement in
inference time over the lifetime of the predictor. I will present
different amortized inference algorithms, which can re-use computation
from previously solved instances to speed up any standard inference
algorithm and retain their exactness properties. Then, I will discuss
an approach for decomposed amortized inference that allows us to apply
this idea for sub-parts of the output structure. Through the talk, I
will present results which show that, for the tasks of semantic role
labeling and entity-relation extraction, the various amortized
inference algorithms give significant decrease in the number of calls
to the underlying inference engine.
I am a post-doc at Stanford University with Chris Manning and the
Stanford NLP group. I obtained my Ph.D. from the Computer Science
Department of the University of Illinois at Urbana-Champaign, where I
worked with Dan Roth and the Cognitive Computation group.
My research deals with Machine Learning and Natural Language Processing.
Please arrive at least 10 minutes early in order to sign in and be escorted to the conference room. SRI is located at 333 Ravenswood Avenue in Menlo Park. Visitors may park in the visitors lot in front of Building E, and should follow the instructions by the lobby phone to be escorted to the meeting room. Detailed directions to SRI, as well as maps, are available from the Visiting AIC web page.
©2014 SRI International 333 Ravenswood Avenue, Menlo Park, CA 94025-3493