The encoded sequence representation in multilayer networks

The author presents an encoded sequence approach to represent output vectors in multilayer networks. The new method is applied to a multilayer network with one hidden layer to classify the letters A to J. Since patterns are represented as sequences rather than individual node outputs, the circular Hamming distance (CHD) operator is proposed to prevent cluster deception at the time of classification. To classify the individual bits of the sequence, a two-way confidence region is proposed based on the statistical estimator of parameters for random data. The results show that when the CHD operator is employed, a reduction of 45 percent, on average, in the number of training iterations and a reduction of 40 percent, on average in the number of required weights are achieved when compared to a network with the conventional one-node vector