Dynamic Sparse Training of Diagonally Sparse Networks | Read Paper on Bytez