[Colloquium] Guest Speaker announcement

[Ponda Barnes]

Guest Speaker
 
Speaker: Vladimir Pavlovic
Speaker's home page: http://www.cs.rutgers.edu/~vladimir
 
Date: Friday, April 20, 2007
Time: 2:00
Location: TTI-C Conference room
 
Title:
 
Discriminative learning methods for state space models
 
Abstract: 
In this talk, I will discuss discriminative learning algorithms for
dynamical systems, commonly known as the state space models. Models such as
Conditional Random Fields or Maximum Entropy Markov Models outperform the
generative Hidden Markov Models in sequence tagging problems in discrete
domains. However, continuous state domains introduce different sets of
constraints that make direct application of discrete state methods
challenging.  We suggest to two solutions to this problem.  In first, we
learn generative state space models with discriminative cost functionals.
For Dynamical Systems, the proposed methods provide significantly lower
prediction error than the standard maximum likelihood estimator, often
comparable to complex nonlinear models.  We then discuss a set of general
conditions that result in convex learning of state space models and propose
a new convex cone algorithm that exploits these conditions. We evaluate the
generalization performance of our methods on the 3D human pose-tracking
problem.  The experiments indicate that the discriminative learning can lead
to improved accuracy of pose estimation with no increase in computational
cost of tracking.
 
If you have any questions or would like to meet the speaker please contact
Ponda Barnes at pondabarnes at tti-c.org.
For future TTI-C talks and event please go to
http://ttic.uchicago.edu/cal/monrh.php
 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20070419/73613005/attachment.html