Faster training of very deep networks via p-norm gates | IEEE Conference Publication | IEEE Xplore