SUPERVISED SELF-CODING IN MULTILAYERED FEEDFORWARD NETWORKS

Supervised neural-network learning algorithms have proven very success ful at solving a variety of learning problems. However, they suffer fr om a common problem of requiring explicit output labels. This requirem ent makes such algorithms implausible as biological models. In this pa per, it is shown that pattern classification can be achieved, in a mul tilayered feedforward neural network, without requiring explicit outpu t labels, by a process of supervised self-coding. The class projection is achieved by optimizing appropriate within-class uniformity, and be tween-class discernibility criteria, The mapping function and the clas s labels are developed together, iteratively using the derived self-co ding backpropagation algorithm. The ability of the self-coding network to generalize on unseen data is also experimentally evaluated on real data sets, and compares favorably with the traditional labeled superv ision with neural networks, However, interesting features emerge out o f the proposed self-coding supervision, which are absent in convention al approaches. The further implications of supervised self-coding with neural networks are also discussed.