[Colloquium] TTI-C Talks Next Week (6/14 & 6/17) Warmuth & Whiteson

[Katherine Cumming]

TOYOTA TECHNOLOGICAL INSTITUTE TALKS
 
 
 
Speaker:  Manfred Warmuth, University of California Santa Cruz
Speaker's homepage:  http://www.cse.ucsc.edu/~manfred/
 
Time:  Tuesday, June 14th, 2005 @ 12:00pm - Lunch Provided by TTI-C
Location:  TTI-C Conference Room 
Title: Leaving the Span
Abstract: 
When linear models are too simple then the following ``kernel trick'' is
commonly used: Embed the instances into a high-dimensional feature space
and use any algorithm whose linear weight vector in feature space is a
linear combination of the expanded instances. Linear models in feature
space are typically non-linear in the original space and seemingly more
powerful. Also dot products can still be computed efficiently via the
use of a kernel function.

However we discuss a simple sparse linear problem that is hard to learn
with any algorithm that uses a linear combination of the embedded
training instances as its weight vector, no matter what embedding is
used. We show that these algorithms are inherently limited by the fact
that after seeing k instances only a weight space of dimension k can be
spanned.

Surprisingly the same problem can be efficiently learned using the
exponentiated gradient (EG) algorithm: Now the component-wise logarithms
of the weights are essentially a linear combination of the training
instances. This algorithm enforces ``additional constraints'' on the
weights (all must be non-negative and sum to one) and in some cases
these constraints alone force the rank of the weight space to grow as
fast as 2^k.

(Joint work with S.V.N. Vishwanathan!) 
 
 
 
Speaker:  Shimon Whiteson, University of Texas at Austin
Speaker's homepage:  http://www.cs.utexas.edu/users/shimon/
 
Time:  Friday, June 17thh, 2005 @ 12:00pm - Lunch Provided by TTI-C
Location:  TTI-C Conference Room 
Title:  <http://ttic.uchicago.edu/events/event_detail.php?event_id=121>
Evolutionary Function Approximation for Reinforcement Learning
Abstract: 
Temporal difference methods are theoretically grounded and empirically
effective methods for addressing sequential decision making problems
with delayed rewards. Most problems of real-world interest require
coupling TD methods with a function approximator to represent the value
function. However, using function approximators requires manually making
crucial representational decisions. This talk will introduce
evolutionary function approximation, a novel approach to automatically
selecting function approximator representations that enable efficient
individual learning. Our method evolves individuals that are better able
to learn. We present a fully implemented instantiation of evolutionary
function approximation which combines NEAT, a neuroevolutionary
optimization technique, with Q-learning and Sarsa, two popular TD
methods. The resulting NEAT+Q and NEAT+Sarsa algorithms automatically
learn effective representations for neural network function
approximators. We evaluate these algorithms with an extended empirical
study in the autonomic computing domain of server job scheduling. The
results demonstrate that evolutionary function approximation can
substantially improve the performance of TD methods. 
 
 
 
-----------------------------------------------------------------
 
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/20050609/1cbf063a/attachment.htm