How Shrinking Gradient Noise Helps the Performance of Neural Networks | IEEE Conference Publication | IEEE Xplore