Training MLPs layer-by-layer with the information potential | IEEE Conference Publication | IEEE Xplore