Dynamics of Batch Learning in Multilayer Networks { Overrealizability and Overtraining {