Submodular neural network is better than modular neural network and support vector machines for personal verification
Takashi Nagano, Makoto Hirahara, Hideaki Eguchi · 2004
A sub-modular neural network (SMNN) proposed a few years ago is compared with the usual modular neural network (MNN) and support vector machines (SVM) in terms of pattern recognition performance. Some computer simulation results showed that SMNN was much superior to MNN and SVM as for rejection rates of patterns in unlearned classes under the condition that they gave almost the same recognition rates for patterns in learned classes. These results strongly suggest that SMNN is more suitable for personal verification systems than the other two as such systems require high rejection rate for patterns in unlearned classes.