Regularize, Expand and Compress: Multi-task based Lifelong Learning via NonExpansive AutoML | Read Paper on Bytez