Specialization in Populations of Artificial Neural Networks.
Henrik Hautop Lund, Brian H. Mayoh · 1995
Specialization in populations of artificial neural networks is studied. Organisms with both fixed and evolvable fitness formulae are placed in isolated and shared environments, and the emerged behaviors are compared. An evolvable fitness formula specifies, that the evaluation measure is let free to evolve, and we obtain co-evolution of the expressed behavior and the individual evolvable fitness formula. In an isolated environment a generalist behavior emerges when organisms have a fixed fitness formula, and a specialist behavior emerges when organisms have individual evolvable fitness formulae. A population diversification analysis shows, that almost all organisms in a population in an isolated environment converge towards the same behavioral strategy, while we find, that competition can act to provide population diversification in populations of organisms in a shared environment.