LIP6 2000/016
- Reports
Un algorithme évolutionnaire adaptatif pour une colonie de fourmis - L. Gacôgne
- 7 pages - 05/31/2000- document en - http://www.lip6.fr/lip6/reports/2000/lip6.2000.016.ps.gz - 69 Ko
- Contact : Louis.Gacogne (at) nulllip6.fr
- Ancien Thème : APA
- Keywords : Foraging problem, Neural networks, Evolutionary algorithms, Coevolution
- Publisher : Valerie.Mangin (at) nulllip6.fr
This paper deals with a simulation of an ant colony which is subject to an evolution. Each one of the ants is moving according to a small neural network, in a circular pitch where it is involved in a colored grid indicating different signals (food, borders, stimuli from other ants . .) They are supposed to look for food and carry it back to their nest located in the center of the playground. But they don't have any rule to do that and we experiment an evolutionary algorithm to select the bests individuals generation to generation.
Face to formalise a comportment as a mathematical function or a rule-based system, we can imagine many ways, so we chose to consider an ant located at any point of the playground with only the knowledge about the five neighboor points front of it. The ant is capable to use then his proper neural network to choose the next case will suit it. The key-point of our evolutionary algorithm is to set up those neural networks. So we study the fitness of each ant and build an offspring in an elitist way, with genetic operators linked with the representation of the networks.
Face to formalise a comportment as a mathematical function or a rule-based system, we can imagine many ways, so we chose to consider an ant located at any point of the playground with only the knowledge about the five neighboor points front of it. The ant is capable to use then his proper neural network to choose the next case will suit it. The key-point of our evolutionary algorithm is to set up those neural networks. So we study the fitness of each ant and build an offspring in an elitist way, with genetic operators linked with the representation of the networks.