[Colloquium] Talks at TTIC: Mehrdad Mahdavi, Michigan State University

[Dawn Ellis]

When:     Thursday, January 23rd at 10am

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

Who:       Mehrdad Mahdavi, Michigan State University

Title:       Exploiting smoothness in statistical learning, online
learning, and
              stochastic optimization

Abstract:

The overarching goal of  this talk is to reassess  the smoothness of loss
functions in online learning, statistical learning, and stochastic
optimization. In particular we  examine how exploiting  smoothness of loss
functions could be beneficial in these settings in terms of sample
complexity, regret analysis,  and convergence rate.

In online setting, we introduce a new measure to characterize the
performance of online learning algorithms which is referred to  as gradual
variation, and devise novel online algorithms  with regret bounded by
gradual variation. The proposed algorithms can take advantage of benign
 sequences and at the same time protect against the adversarial sequences.

In stochastic optimization setting, we propose a novel optimization
paradigm that is referred to as mixed optimization which interpolates
between stochastic and full gradient methods and is able to exploit the
smoothness assumption to obtain sharper or condition number independent
 convergence rates.

Finally, in passive learning framework, we show that  by exploiting the
smoothness of loss functions and with the help of target risk, we are able
to improve the sample complexity exponentially  by  proposing a properly
designed learning algorithm.

Host: Greg Shakhnarovich, Greg 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/20140116/831106f4/attachment.htm