To be held on June 29, 2006, at Carnegie-Mellon University, Pittsburgh, USA as part of the workshop day at ICML 2006.
Every participant to the ICML'2006 conference is welcome to attend and
participate to the workshop. This requires that the fee for the
workshops has been paid. For that, please refer to the ICML
registration page.
On behalf of the presentations of submitted papers and invited
speeches, we want to keep the discussion very open to anyone who
wishes to express his/her opinion, as far as this is relevant with the
topic of the workshop. To let us somehow organize the day, and have a
sort of schedule, we kindly ask anyone who would like to attend to:
Please, send emails to philippe -dot- preux -at- univ-lille3 -dot- fr
Reinforcement Learning (RL) is an adaptive method for learning to make
good decisions in a complex, stochastic and partially unknown
environment.
In order to deal with large-scale RL problems, the functions of
interest (such as the value function or the policy, or a model of the
unknown state dynamics) must be approximately represented. Since the
quality of approximations directly influences the performance measures
of ultimate interest, the function approximation methods employed
should be sample-efficient whilst being able to deliver high quality
estimates at the same time. For instance, in Approximate Dynamic
Programming the performance of policies greedy with respect to
approximate value functions are bounded in terms of the
approximation's precision.
Thus real-world applications of RL need efficient function approximators.
Kernel methods are at the heart of many modern machine learning
techniques. They make it possible to derive efficient algorithms that
work in function spaces of high representation power and come with
PAC-style theoretical results.
This workshop will be entirely dedicated to bridging
the gap between kernel methods and reinforcement learning.
Appropriate topics for papers include, but are not limited to:
This is the temptative schedule of the workshop day (as of June, 28th).
We primarely expect original submissions. However, we will also accept submissions of already accepted papers if the authors make it clear that this is not an original submission. In the process of acceptence, we will favor original submissions and accept resubmissions only if there is enough time in the schedule.