符号函数激励的WASD神经网络与XOR应用

Signum-Function-Activated WASD Neuronet and Its XOR Application

基于权值与结构确定(WASD)算法,提出和构建了一种以非连续符号函数为隐层神经元激励函数的WASD神经网络模型。通过WASD算法,能有效地确定所构建网络的权值及网络的最优结构。该文也将此网络模型应用于XOR(异或)上,并详细讨论了在带噪类型不同时网络在此应用上的性能。计算机数值实验结果验证了所提出的权值与结构确定法能够有效地确定出网络的最优权值与结构,所构建的WASD网络在XOR应用上具有优秀的抗噪性能。另外,通过对比符号函数激励的WASD神经网络与幂函数激励的WASD神经网络在高维XOR应用方面的性能差异,证实了所提出的符号函数激励的WASD神经网络及算法在解决非线性问题时的优越性。

A discontinuous signum-function-activated （SFA） weights-and-structure-determination （WASD） neuronet model is presented and constructed based on the WASD algorithm. By this algorithm, the optimal weights and structure can be determined effectively. We apply the SFA-WASD neuronet mod- el to XOR （ i. e. , exclusive or） , and detail its performance in the XOR application with various types of disturbance noise considered. Numerical verification results substantiate the validity of the WASD algo- rithm in determining the optimal weights and structure, as well as the good anti-noise ability of the SFA- WASD neuronet in the XOR application. Moreover, for high-dimension XOR application, the perform- ance comparison is made between the power-function-activated （PFA） WASD neuronet and the SFA- WASD neuronet. The numerical results verify the superiority of the SFA-WASD neuronet in terms of solving nonlinear problems.