Optimize-and-Learn: a Parameter Tuning Framework applied for the Divide-and-Evolve Method used to AI PLanning

In this paper we describe the system used in the Plan-ning and Learning Part of the 7th International Plan-ning Competition. Learn-and-Optimize (LaO) is a generic surrogate based method for parameter tuning combining learning and optimization. In this paper LaO is used to tune Divide-and-Evolve (DaE), an Evo-lutionary Algorithm for AI Planning. The LaO frame-work makes it possible to learn the relation between some features describing a given instance and the op-timal parameters for this instance, thus it enables to extrapolate this relation to unknown instances in the same domain. Moreover, the learned model is used as a surrogate-model to accelerate the search for the op-timal parameters. It hence becomes possible to solve intra-domain and extra-domain generalization in a sin-gle framework. The proposed implementation of LaO uses an Artificial Neural Network for learning the map-ping between features and optimal parameters, and the Covariance Matrix Adaptation Evolution Strategy for optimization