Separability Conditions for Multilayer Nets Having Solutions and Convergent Superiority of Bipolar Nets
Hiroshi Shiratsuchi, Hiromu Gotanda, Katsuhiro Inoue, Kousuke Kumamaru · Journal of Advanced Computational Intelligence and Intelligent Informatics · 2004
Separability conditions are formulated for multilayer nets to have solutions by a set of normal vectors orthogonal to separation hyperplanes. Comparing separability conditions to distributions of normal vectors with weights and biases initialized ordinarily by random numbers with a zero mean, we found that bipolar nets are superior to unipolar nets in convergence of the back propagation learning initialized in such an ordinary manner.