Gravar-mail: Vitality of Neural Networks under Reoccurring Catastrophic Failures