Abstract
In this study, an attempt is made to develop a new pattern recognition system by introducing neural network using interpolating vector method. In the interpolating vector method, reference vectors with labels are generated through the conflict learning in the multi-dimensional state space. Then interpolating vectors are formed on the line connecting all pairs of the reference vectors with the same label. From all the interpolating vectors, a vector closest to the test vector is chosen, whose label provides the result of
the pattern recognition. In this study, it is assumed that the multi-dimensional state space is constructed by the intermediate layer obtained through the learning by minimizing the error. Furthermore, a method
using extrapolating vectors is developed for the case with learning difficulty, in order to improved the accuracy of pattern recognition. Through several numerical examples with actual data, it is shown that the proposed method is useful to improve the recognition rate.
