Mitigating fooling with competitive overcomplete output layer neural networks | IEEE Conference Publication | IEEE Xplore