Nesterov momentum is a variant of the momentum algorithm that differs from the momentum method only at the point the gradient is calculated. The standard momentum method computes the gradient first at the current location and then takes a big jump in the direction of the accumulated gradient. The Nesterov momentum first makes a big jump along the direction of the previously accumulated gradient and then computes the gradient at the new point. The new gradient is corrected by again taking the EWMA of all previous gradients: