Continual Learning With Quasi-Newton Methods