Going beyond Classification Problems for the Continual Learning of Deep Neural Networks