基于累积量的两层前馈神经网络盲辨识

由于非线性系统输出是其参数的非线性函数,直接利用高阶累积量辨识两层前馈神经网络(FNN)通常是十分困难的。为解决这一问题,该文提出两种基于四阶累积量的FNN辨识方法。第一种方法,FNN的隐元在其输入空间利用多个线性系统近似,进而FNN利用一统计模型混合专家(ME)网络重新描述。基于ME模型,FNN参数可利用统计期望值最大化(EM)算法获得估计。第二种方法,为简化FNN的ME模型,引入隐含观测量。基于隐含观测量估计,FNN被分解为多个单隐元的训练问题,进而整体FNN可利用一两阶层ME描述。基于单隐元的参数估计,FNN可利用一具有更快收敛速度的简化算法获得估计。

Cumulant-bansed blind identification of two-layer feedforward neural networks

It is always difficult to train a two-layer Feedforward Neural Networks (FNN) based on the cunmlant match criterion because cumulants are nonlinear implicit functions of the FNN parameters. In this work, two new cumulant-based training methods for two-layer FNN are de-veloped. In the first method, the hidden units of two-layer FNN are approximated with multiple linear systems, and further total FNN is modeled with a "mixture of experts" (ME) architec-ture. With the ME modei, FNN parameters are estimated with expectation-maximization (EM) algorithm. The second method, for simplifying the two-layer FNN statistical modei, proposes a, simplified two-level hierarchical ME to remodel the FNN, in which hidden variables are in-troduced such that training total FNN is changed to training a set of single neurons. Based on training single neuron, total FNN is estimated in a simplified version with a faster convergence speed.