Analysis of Feature Extraction Criterion Function Maximum in Nonlinear Multi-layer Feedforward Neural Networks for Pattern Recognition
Cao Junhong, Wei Zhuobin, Huang Tao, Xianwei Xiong · 2010
This paper addresses feature extraction criterion function of multi-layer feed forward neural networks with linear output units and nonlinear hidden units. From the minimum mean square error function of the network output, the paper uses the nature of matrix trace and singular value decomposition, deduces the formula for calculating the nonlinear criterion function maximum, then explains the significance of this formula. Finally, simulation examples prove the correctness of the analytic style.