Thank you for your interest in this research. We are investigating the effect of an evolution rejection strategy on the performance of interactive genetic algorithms.

Instructions

The aim is to evolve a colour that matches the target colour as closely as possible using human interaction. This is done by selecting the closest colour match from a swatch of nine candidate colours. The algorithm will then create a new swatch of colours from which the best match to the target must once again be selected. Additionally, at each stage you can guide the process more effectively by rejecting colours that are poor matches to the target colour. This works best if you only reject a few dissimilar colours to the target. When you feel that you have achieved a good colour match and no more improvement can be made, select the best match from the current colour swatch and then click finish.

To start the experiment click on the link below which will launch a Java applet.

Online Interactive Genetic Algorithm Experiment

This study is based on work by Breukelaer et al (1).

Note: The applet requires Java 1.4.2

About Genetic Algorithms

Evolutionary algorithms are search techniques inspired by evolutionary biology and employing operations such as inheritance, mutation, selection and crossover. Evolution starts from a population of randomly generated individuals and occurs over a number of generations. At each generation, the individuals that best solve the problem are selected and recombined to form new individuals that can represent better solutions.

In an interactive evolutionary algorithm the user cannot be expected to rate a large set of individuals at each generation, so the selection task has to be kept simple. This is done by using a small number of individuals at each generation and asking the user to simply select the best. The best (or fittest) individual is then cloned and mutated to form a new population. By also allowing the user to reject some of the individuals it is hoped that this extra information can be used to increase the speed at which the algorithm converges.

Contact

For further information, contact Ben George - bg37@kent.ac.uk

References

2) M. Herdy. Evolution Strategies with Subjection Selection. Parallel Problem Solving from Nature IV, Berlin, Germany, 1996.