Promoting social diversity for the automated learning of complex MDE artifacts

Software modeling activities typically involve a tedious and time-consuming effort by specially trained personnel. This lack of automation hampers the adoption of model-driven engineering (MDE). Nevertheless, in the recent years, much research work has been dedicated to learn executable MDE artifacts instead of writing them manually. In this context, mono- and multi-objective genetic programming (GP) has proven being an efficient and reliable method to derive automation knowledge by using, as training data, a set of examples representing the expected behavior of an artifact. Generally, conformance to the training example set is the main objective to lead the learning process. Yet, single fitness peak, or local optima deadlock, a common challenge in GP, hinders the application of GP to MDE. In this paper, we propose a strategy to promote populations’ social diversity during the GP learning process. We evaluate our approach with an empirical study featuring the case of learning well-formedness rules in MDE with a multi-objective genetic programming algorithm. Our evaluation shows that integration of social diversity leads to more efficient search, faster convergence, and more generalizable results. Moreover, when the social diversity is used as crowding distance, this convergence is uniform through a hundred of runs despite the probabilistic nature of GP. It also shows that genotypic diversity strategies cannot achieve comparable results.