Spartan: A Sparsity-Adaptive Framework to Accelerate Deep Neural Network Training on GPUs | IEEE Journals & Magazine | IEEE Xplore