UTCS Corporate Connection/FoCS/Data Mining: "Learning with Exploration", John Langford/ Yahoo, ENS 115, March 24, 2011, 11:00 a.m.

Contact Name: 
Jenna Whitney
Mar 24, 2011 11:00am - 12:00pm

Type of Talk: UTCS Colloquia/FoCS/Data Mining


liation: John Langford/Yahoo

Date/Time: Thursday, March 24, 2011 11


Location: ENS 115

Host: FoCS

Talk Title: Learning with Ex


Talk Abstract: The default successful paradigm of machine le

arning is supervised learning. Here humans are hired (often through mechan

ical turk these days) to label items, and then the labeled information is

fed into a learning algorithm to create a learned predictor. A more natura

l, less expensive, and accurate approach is observe what works in practic

e, and use this information to learn a predictor.

For people famili

ar with the first approach, there are several failure modes ranging from p

lain inconsistency to mere suboptimality. A core basic issue here is the

need for exploration---because if a choice is not explored, we can''t opti

mize for it. The need for exploration implies a need for using exploration
to evaluate learned solutions, to guide the optimization of learned predi

ctors, and the need to control the process of exploration so as to accompl

ish it efficiently.

I will provide an overview of the problems which c

rop up in this area, and how to solve them. This includes new results on

policy evaluation and optimization, with the first ever optimization-based
exploration control algorithms for this setting.

Speaker Bio: Dr. Joh

n Langford is a Senior Researcher at Yahoo! Research. His work includes res

earch in machine learning, game theory, steganography, and Captchas. He

was previously a Research Associate Professor at the Toyota Technological I

nstitute in Chicago. He has worked in the past at IBM''s Watson Research Ce

nter in Yortown, NY under the Goldstine Fellowship. He earned a PhD in com

puter science from Carnegie Mellon University in 2002 and a Physics/Compute

r Science double major from CalTech in 1997.