层次式级联多网络模型及其性能

A Layered Cascade Multi Network for Pattern Classification

本文提出了一种层次式级联多网络结构，论述了其训练算法，并验证了该结构的收敛性。这种结构在处理高阶非线性分布数据的分类问题方面具有优良的性能，用北京谱仪事例过滤进行性能测试结果表明，其训练和泛化性能都优于单个ＢＰ网络。

A layered cascade multi network architecture and learning algorithm have been proposed in this paper. It has been proved that zero classification errors can be reached for any finite, non contradictory training sets in multi category classification tasks. Such architecture is more suitable for those classification tasks having high order non linear distribution. Finally, we tested the network by the two spiral problem and BES data.