Regularizing Second-Order Influences for Continual Learning | Read Paper on Bytez