Rules

1. Each entry will be run on a set of 30 problems. The problems will be drawn from three problem series in the 2D Fractal Function benchmarking suite, with 10 problems drawn randomly from each series.
2. The three problem series to be used in the competition will be specified and made available to contestants in advance so that contestants can hone their entries for optimal performance. Examples (subject to change) are provided in the pre-release code, and will be fixed by the opening date of the competition (see Important Dates). 
3. The actual problem instances used in the competition will be randomly drawn from the three problem series and will be unknown to contestants (to prevent over-specialisation). The code that is used to construct the problem set and test the entries is freely available prior to the start of the competition.
4. Within each problem, the starting position and scale are generated randomly (but the same for each entry). 
5. For each problem, the contestant's algorithm will be allocated 1000 evaluations. The algorithm may use these evaluations in any way the designer sees fit (see the note below about the "Spirit of the Rules"). For example, evolutionary, swarm, local search, hybrids, self-adaptive and learning algorithms are encouraged to compete on an even playing-field.
6. The competition software will record the best (minimum) value found by each entry on each of the 30 problems. The entries will then be ranked on each problem according to the minima achieved. This will give a score (ranking) for each entry on each problem.
7. The winning entry will be the one with the best median score across all 30 problems. The winner will be asked to give a short presentation on their algorithm at the Competitions Session at WCCI2008. If the schedule permits the runner-up may also be invited to present their algorithm.

Prize

The winner will receive a free registration for WCCI2008.

Submission

Entries should be emailed to  This e-mail address is being protected from spambots. You need JavaScript enabled to view it . Details of the methods that must be called by your code can be found in the Downloads. Entries are currently permitted in Java and Matlab. For other languages please contact the organiser.

Updates

Contestants should check back at this site periodically (or join the email list) for any updates or clarifications to the rules. 

Spirit of the Rules 

The goal of the competition is to find which generic techniques best handle the problem of exploration at unknown scale across a variety of problems. Since we are unfortunately limited to 64-bit double precision, the fractal functions ultimately "bottom out", and with a little knowledge of the IEEE floating point standard, it is conceivable (tho of unknown net benefit) to use some of the 1000 evaluations in a hard-wired routine to attempt to guess the scale. While we are happy to hear of the success or otherwise of such attempts, since they do not advance the field they will not be eligible for the prize. Genuine self-adaptive or learning techniques that can be generically applied are of course welcome.