A Penalty-Function Approach for Pruning Feedforward Neural Networks | MIT Press Journals & Magazine | IEEE Xplore