GECCO 2008 Contest Problems

Here are the three contest problems for 2008 and links to all you will need to do them. This year we want to push the bounds of what is evolved by introducing some novel problems in fringe areas. These problems are suitable for any EC practitioner. We welcome submissions from students in EC classes.

Deadline: June 27 at 5PM PDT (note the timezone)

A 2D Packing Problem

This problem is a 2D variation of bin packing problems, which requires new ways to evolve with a 2D chromosome. The goal is to best pack a grid to maximize the sum of scores where every unique pair of adjacent numbers in the grid has its own score. [The details]

Entrants must turn in their best grid and a brief summary of their evolutionary algorithm. The winner is based on the best score and the quality of the evolutionary algorithm.

Human Evaluation of Evolved L-System Images

The goal is to evolve an L-System that will recreate a series of images in a defined number of cycles. Entrants must turn in their L-System grammar, the series of images the L-System generates and a brief summary of their evolutionary algorithm. You are provided with restrictions on the L-system grammar, target images, and a function to generate the series of images for any valid L-System. [The details]

The submitted grammars will be run on the provided L-system generator and compared by human inspection for how close they are precevied to replicate features of the exected image. The judges may or may not know what an L-system is. A human panel will judge the quality of the images.

Finding a Balanced Diet in Fractal World

In a more complex variant of the Santa Fe trail, the goal is to evolve an agent to search a landscape and find as much as possible of two types of food. The landscape is a fractal with varying elevations and impassible regions. Problem details, including sample training maps. The requirements for the function, including arguments and outputs, are defined through the details link. [The details]

Entrants must turn in an ANSI C function representing the best evolved agent and a brief summary of their evolutionary algorithm. Scoring will be based on a fitness function that takes into account not only the amount of the two types of food collected but the balance between the two types of food. Submitted functions will be tested on a number of random, test maps. The quality of the evolutionary algorithm may be taken into account in the scoring.

Submission Details

Submissions will be made via the submission page here.

Questions about these pages or the problems themselves can be sent to [Terry Soule] or [Robert Heckendorn]