[Colloquium] Talks at TTIC: Kwang-Sung Jun, Univ of Wisconsin-Madison

[Dawn Ellis]

When:     Friday, March 6, 2015 at 10am

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

Who:      Kwang-Sung Jun, Univ of Wisconsin-Madison

Title:      Initial-Visit Emitting Random Walk

Abstract:
I am going to talk about a new random walk model that has an interesting
application in cognitive science. Imagine a random walk that outputs a
state only when visiting it for the first time. The observed output is
therefore a censored version of the underlying walk, and consists of a
permutation of the states or a prefix of it. I call this model
initial-visit emitting random walk (INVITE). INVITE provides a unique
computational model for certain physical processes. However, learning in
INVITE is challenging because naive computation of the likelihood by
marginalizing over infinitely many hidden random walk trajectories is
intractable. In this talk, I propose the first efficient maximum likelihood
estimate (MLE) for INVITE by decomposing the censored output into a series
of absorbing random walks. I also present theoretical properties of the MLE
including identifiability and consistency. I then turn to an important
application of INVITE in cognitive science to study human memory search. I
show that INVITE outperforms several existing methods on this application.

Bio:
Kwang-Sung Jun is a Ph.D. candidate in Department of Computer Sciences at
University of Wisconsin-Madison, advised by Prof. Xiaojin Zhu. His research
focuses on new machine learning models motivated by cognitive psychology
and pure-exploration multi-armed bandit problems.

Host:  Nati Srebro, nati 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/20150303/ab204499/attachment.htm