stochastic gradient algorithms have been the main focus of large-scale learning problems and they led to important successes in deep learning. Convergence of SGD depends on how carefully learning rate of the algorithm is tuned and the noise in stochastic estimates of the gradient. In t