Co-evolution and ecosystem based problem solving

Abstract

Emergent cooperative relations in ecosystems are ill understood, but have the potential to strongly improve evolutionary computing. On the other hand, eco-evolutionary computation has the potential to provide new insights in the structuring and functioning of ecosystems. Here we study ecosystem based problem solving in a co-evolutionary framework of predators (solvers) and prey (problems), extended with a population of scavengers, which can eat the remains of prey (that is, cooperate with the predators in solving the problems). We show that such an artificial ecosystem of predators, prey and scavengers, with a selection and fitness regime favoring specialization, self-organizes in space and time such that (1) problems are automatically decomposed in easier to solve parts, (2) the predator, prey and scavenger populations differentiate in sub-populations according to this decomposition, and (3) predators and scavengers automatically co-localize in space such that the problems are indeed solved by predator–scavenger combinations which together correctly approximate the target function. That is, the use of a spatial co-evolutionary ecosystem as information processing unit for evolutionary computation gives rise to an emergent structure of niches, each consisting of complementary partial solutions. As a result, ecosystem based solutions are preferred over individual-based solutions in solving the studied function approximation task.