The 2014 competition


We are pleased to announce the 8th reinforcement learning competition (2014). As in previous editions we invite RL researchers and students to compete in a variety of problem domains. Not only is this an opportunity to test your algorithms on carefully constructed problem settings, and compare to other state-of-the-art algorithms, but also is it also a challenge to create new algorithms that perform well at problems with different characteristics. Last but not least, we hope to advance our understanding of RL problems and algorithms by identifying what makes a learning algorithm good at what types of problems.

Call for volunteers and domains

We are looking for help with setting up and maintaining the server.
We are also looking for new domain suggestions, to potentially replace older ones. Moreover, we encourage submission of domains which allow to test algorithms on: 1) Multi-task problems and task transfer, 2) Adversarial problems. 

Preliminary planning

  • The competition server opens: October, 2014.
  • The competition: November, 2014.
  • The workshop: To be determined.

NB: these dates may change.

Workshop

At the end of the competition authors are requested to submit a paper to the competition workshop, co-located with a major conference. At this workshop, authors present their work, and discussions will be held about the characteristics of the problems and the algorithms in relation to the performance of the agents.

Competition 
The competition will be hosted at http://www.rl-competition.org/

Please have a look at the domains, and take up the challenge. Agents may compete in one or more domains. For specific domains, such as the helicopter and invasive species, agents are tested on variations of a general problem class. The 'polyathlon' domain, is more general. There agents act in completely a completely unknown set of environmetns.

Organizers

  1. Christos Dimitrakakis
  2. Nikolaos Tziortziotis