Improving rule extraction from neural networks by modifying hidden layer representations | IEEE Conference Publication | IEEE Xplore