Theory-Laden Design of Mutation-Based Geometric Semantic Genetic Programming for Learning Classification Trees

is a recently introduced form of Genetic Programming (GP), rooted in a geometric theory of representations, that searches the semantic space of functions. The fitness landscape seen by GSGP is always – for any domain and for any problem – unimodal with a linear slope by construction. This makes the search for the optimum much easier than for traditional GP, and it opens the way to analyse theoretically in a easy manner the optimisation time of GSGP in a general setting. Very recent work proposed a runtime analysis of mutation-based GSGP on the class of all Boolean function learning problems. We present a runtime analysis of mutation-based GSGP on the class of all classification tree learning problems, which is a classical application domain of standard GP. I