A hyperellipsoid neural network for pattern classification
Chang Jou, Quen Zong Wu, Shuh‐Chuan Tsay, Yuh-Jiuan Tsay, Shih-Shien Yu · 1991
Proposes a distance based neural network for pattern classification. In the beginning of network training, a hyperellipsoid is constructed for each training pattern. The authors try to merge the hyperellipsoids of the same classes without interfering with the hyperellipsoids of other classes. Because each class is represented by several hyperellipsoids, any pattern which is located in or nearest to one of these hyperellipsoids is classified to this class. This distance based neural network does not need any hidden layer. An example is given to compare the performances of this network and the multilayer perceptron. It shows that better performance may be obtained by using this network.>