基于卡尔曼滤波的二次型神经网络学习算法及收敛性分析

提出了二次型多层前馈神经网络的卡尔曼滤波学习算法，并证明了该算法的收敛性。与文献［２，３］中的学习算法和经典的误差反向传播学习算法相比，新的学习算法具有更快的学习速度、良好的泛化能力，并且对学习率有很好的鲁棒性，不容易陷入局部极小点。仿真实验结果表明了新算法的有效性。

A Kalman Filter Learning Algorithm for Multilayer Quadratic Neural Networks and Its Convergence Analysis

This paper analyzes the multilayer quadratic neural networks and a kind of Kalman filter learning algorithm.Then a new Kalman filter learning algorithm for the quadratic neural networks is presented and the convergence of the algorithm is proved.Experiment results show that the new algorithm can speed learning of the feedforward quadratic neural networks and is robust against the variation of learning rate.The proposed algorithm can also achieve good generalization capability.