USE OF A QUASI-NEWTON METHOD IN A FEEDFORWARD NEURAL-NETWORK CONSTRUCTION ALGORITHM

Interest in algorithms which dynamically construct neural networks has been growing in recent years. This paper describes an algorithm for c onstructing a single hidden layer feedforward neural network. A distin guishing feature of this algorithm is that it uses the quasi-Newton me thod to minimize the sequence of error functions associated with the g rowing network. Experimental results indicate that the algorithm if ve ry efficient and robust. The algorithm was tested on two test problems . The first was the n-bit parity problem and the second was the breast cancer diagnosis problem from the University of Wisconsin Hospitals. For the n-bit parity problem, the algorithm was able to construct neur al network having less than n hidden units that solved the problem for n = 4,...,7. For the cancer diagnosis problem, the neural networks co nstructed by the algorithm had small number of hidden units and high a ccuracy rates on both the training data and the testing data.