Connectionists: 2013 Reinforcement Learning Competition and ICML Workshop

[Guangliang Li]

[We apologize in advance if you receive multiple copies of this message]
----------------------------------------------------------------------------------------------

2013 Reinforcement Learning Competition and ICML Workshop

 20-21 June 2013 Atlanta, USA

https://sites.google.com/site/rlcomp2013/icml_workshop

* *

Competition goal

After a four year hiatus, the reinforcement learning competition is back.
The primary aim of the competition is to test both general
and domain-specific reinforcement learning algorithms, using an
unbiased and transparent methodology. As a side-effect, the competition
will generate a set of benchmarks domains and benchmark results in
those domains, which can then be used as a basis of comparison in
future work.

In the reinforcement learning competition <http://2013.rl-competition.org/>,
researchers can test their algorithms and insights in a friendly
competitive way, on new and challenging domains. The reinforcement learning
competition has not been organized for a while, but will take place again
this year after a long hiatus. The primary aim of the competition is to
test both general and domain-specific reinforcement learning algorithms,
using an unbiased and transparent methodology. The domains used in the
competition will form a set of benchmarks, and the results of the submitted
algorithms will form a body of benchmark results, which researchers can
then use as a basis of comparison in future work.

A secondary aim is to discuss methodological approaches for comparing
reinforcement learning algorithms. This remains an issue in reinforcement
learning in general.

Another important aim is to ensure the continuing existence of the
competition, and the prevention of further hiatuses. Our aim is to ensure
that the competition will organized annually again, from this year onward.
To this extend we will try to build up an organization of interested
researchers.

As in the previous years, the competition will be hosted at

http://www.rl-competition.org/

The format will be remain as is. Agents may compete in one or more of a set
of known domains. This set will include a polyathlon event, where the
agents compete in a sequence of arbitrary environments.

We envisage the actual competition to take place in *June 2013*.



ICML Workshop on the Reinforcement Learning Competition 2013 (WRLCOMP)

The competition is associated with an ICML Workshop in conjunction with ICML
2013 <http://icml.cc/2013/>. The ICML reinforcement learning competition
workshop will bring together researchers, who participated in the
competition, in order to present and discuss their results. We will
evaluate what works, and also under what conditions established methods may
not work so well. In this way we hope to broaden our insight into
state-of-the-art RL algorithms, and important properties of RL problems.

The workshop will be organized along the lines of the domains. One time
slot will be reserved for each competition category winner. In addition,
all competition entrants will be invited to submit a short paper describing
their approach.


We will reserve a special time slot for papers on methodological problems
that arise when comparing reinforcement learning algorithms.

Workshop website: https://sites.google.com/site/rlcomp2013/icml_workshop**

Topics

1. For entrants

All entrants are additionally invited to submit a paper describing the
approach used in the competition, including findings and ideas that relate
to more general research questions.

2. For everyone

You are invited to submit a paper discussing methodological issue when
comparing reinforcement learning algorithms, including but not limited to
the following topics:

Effects of PRNGs in large-scale comparisons

Experimental design for an unbiased evaluation

Metrics for performance evaluation

Reproducibility of results

Robustness of algorithms

Statistical testing standards for reinforcement learning



Submission



Interested authors should format their papers according to ICML formatting
guidelines, available here:
http://icml.cc/2013/wp-content/uploads/2012/12/icml2013stylefiles.tar.gz

Papers should not exceed 4 pages and are due by May 15, 2013. All papers
must be submitted as PDF, and be made through Easychair at *
https://www.easychair.org/conferences/?conf=wrlcomp2013*

Important Dates



Domains available:  April 1

Testing starts: May 1

Competition:  June 1

Paper Submission Deadline: May 15, 2013

Author Notification: June 1, 2013

Workshop: June 20-21, 2013
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.srv.cs.cmu.edu/mailman/private/connectionists/attachments/20130318/7b2ea82f/attachment-0001.html>