[Colloquium] TTI-C Talks This Week: Tarres, Binev, Dahmen, Pontil, Wieder, Rakhlin (4/5-4/8)

[Katherine Cumming]

TOYOTA TECHNOLOGICAL INSTITUTE TALKS
 
Learning Theory Program (1)
 
Speaker:  Pierre Tarres, University of Paul-Sabatier
Speaker's homepage:  http://www.lsp.ups-tlse.fr/Fp/Tarres/
Time:  Tuesday, April 5th @ 3:00pm
Location:  TTI-C Conference Room 
Title: Stochastic Algorithms and Learning
Abstract:
We discuss the asymptotic behavior of some stochastic algorithms arising
in learning and self-organization. First we introduce some new results
on the two-armed bandit problem, interpret them in terms of stochastic
approximation of an underlying dynamical system, and explain how the
methods in this framework adapt to other questions in reinforcement
learning in games and generalized urn models. Then we present some
problems related to self-organization appearing in reinforcement on
graphs, whose study requires a combination of stochastic approximation
and Markov chain methods. 
 
Learning Theory Program (2)
 
Speaker:  Peter Binev, University of South Carolina
Speaker's homepage:  http://www.math.sc.edu/~binev/
Time:  Wednesday, April 6th @ 10:00 am
 
Title:  Learning a Surface
 
 
 
Learning Theory Program (3)
 
Speaker:  Wolfgang Dahmen, Aachen-Germany
Speaker's homepage:  http://elc2.igpm.rwth-aachen.de/~dahmen/
Time:  Wednesday, April 6th @ 3:00pm
 
Title:  Learning with Piecewise Polynomials
 
 
 
Learning Theory Program (4)
 
Speaker:  Massi Pontil, University College-London
Speaker's homepage:  http://www.cs.ucl.ac.uk/staff/M.Pontil/
Time:  Thursday, April 7th @ 10:00 am
 
Title:  Multi Task Learning
 
 
 
Guest Speaker (1)
 
Speaker:  Udi Wieder, Weizmann Institute of Science-Rehovot Israel
Speaker's homepage:  http://www.wisdom.weizmann.ac.il/~uwieder/
Time:  Thursday, April 7th@ 3:00pm
Title:  Dynamic and Scalable Data Structures
Abstract:
A Distributed Hash Table (DHT) is a giant hash table that is
cooperatively maintained by a large number of machines worldwide. The
machines join and leave the system autonomously. The large scale and
dynamism of the system calls for novel design techniques which emphasize
decentralization and automatic re-configuration.

We present several constructions of DHT's and discuss the main design
principles common to many of them. We will focus on the Skip-Graph
structure suggested by Aspnes et al. and Harvey et al. We show a new
routing scheme for Skip-Graphs called Neighbor of Neighbor, through
which Skip-Graphs achieve optimal degree/path-length tradeoff. We also
show that Skip Graphs are expanders (with high probability), and
demonstrate possible benefits of this property to uniform sampling and
load balancing. 
 
Machine Learning Seminar (1)
Speaker:  Alexander Rakhlin, MIT
Speaker's homepage:  http://cbcl.mit.edu/people/rakhlin/
Time:  Friday, April 8th @ 12:15pm
 
Title:  TBA
 
 
-----------------------------------------------------------------
 
If you have questions, or would like to meet the speaker, please contact
Katherine at 773-834-1994 or kcumming at tti-c.org.   For information on
future TTI-C talks and events, please go to the TTI-C Events page:
http://www.tti-c.org/events.html.  TTI-C (1427 East 60th Street,
Chicago, IL  60637)
 
 
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20050405/fdc3aefa/attachment.htm