函数链神经网络的性能改进

PERFORMANCE IMPROVEMENTS OF FUNCTIONAL LINK NEURAL NETWORK

函数链网络(Functional Link Network——FLN)通过对输入向量(或模式)的非线性扩展,将非线性映射特性引入了单层神经网络,采用δ学习规则获得了快速的学习和非线性映射特性。本文在FLN基础上,借助凸集优化思想,利用最陡梯度下降技术获得了比FLN更高的存储容量和更快速的学习速度。计算机模拟的结果证实了所提的算法性能。

Functional Link Network—— FLN introduces the nonlinear mapping ability to single-layer neural network via the nonlinear expansion for input vectors(or patterns)and gains fast learning and nonlinear mapping characteristics using 8 learning rule. In this paper,based on FLN and convex set optimization , we obtain more fast learning with steepest gradient descent method. A computer simulation shows the performance of the proposed method.