Genetic programming theory

Genetic Programming (GP) is a complex adaptive system with an immens number of degrees of freedom. Understanding how, why and when it work is difficult. Its behaviour is typically investigated in two ways experimentally and theoretically. Experimental studies require the experimenter to choose which problems, parameter settings and descriptors to use. Plotting the wrong data increases the confusion about GP's behaviour, rather that clarify it. A more powerful alternative is to study GP theoretically. In this tutorial I will look at GP as a search process and explain its behaviour by considering the GP search space, in terms of its size in its limiting fitness distributions and also the halting probability. I will then use modern schema theory to characterise GP search. Finally, I will be in a position to explain the reasons for spurious phenomena (such as bloat) in GP and I will look into theoretically-sound ways of curing them. Some prior knowledge of GP will be assumed