Dynamics of Co-evolutionary Learning

Abstract

Co-evolutionary learning, which involves the embedding of adaptive learning agents in a fitness environment which dynamically responds to their progress, is a potential solution for many technological chicken and egg problems, and is at the heart of several recent and surprising successes, such as Sim's artificial robot and Tesauro's backgammon player. We recently solved the two spirals problem, a difficult neural network benchmark classification problem, using the genetic programming primitives set up by [Koza, 1992]. Instead of using absolute fitness, we use a relative fitness [Angeline & Pollack, 1993] based on a competition for coverage of the data set. As the population reproduces, the fitness function driving the selection changes, and subproblem niches are opened, rather than crowded out. The solutions found by our method have a symbiotic structure which suggests that by holding niches open, crossover is better able to discover modular building blocks.