TensorDash: Exploiting Sparsity to Accelerate Deep Neural Network Training | IEEE Conference Publication | IEEE Xplore