Machine Learning for a General Purpose Declarative Scene Modeller
Dimitri Plemenos, Georges Miaoulis, Nikos Vassilas · 2002
In this paper we discuss about the implementation of machine learning mechanisms in declarative scene modelling. After a study of the different kinds of declarative modellers and the different cases where machine learning seems useful, we describe two implemented techniques allowing machine learning for declarative modelling by hierarchical decomposition. The first technique is based on neural networks and allows reduction of the solution space in order to generate only solutions corresponding to the user’s wishes. The second one uses a genetic algorithm which, starting from a set of scenes produced by the generation engine of the declarative modeller, produces other solutions under the user’s control, taking hence the place of the generation engine. The obtained results are then explained and discussed.