An optimized backpropagation with minimum norm weights

A backpropagation learning algorithm is presented. The algorithm is a combination of the conventional backpropagation and an objective of minimizing the norm of weights. It is optimal in the sense that it can learn to achieve a set of minimum norm weights while still possessing the best error performance. Fast learning is proven in the algorithm. Simulation results strongly prove its good prospects. The uniqueness of the norm of weights is also demonstrated in the simulation. This algorithm is actually an example of a class of optimized back-propagation learning. The generalization for some problems is straightforward