Strategy Learning with Multilayer Connectionist Representations