Relearning after Damage in Connectionist Networks: Implications for Patient Rehabilitation