CEST: Computation-Efficient N:M Sparse Training for Deep Neural Networks | IEEE Conference Publication | IEEE Xplore