Accelerating Deep Learning via Importance Sampling | Random Notes