一种序贯学习神经网络及其应用

A SEQUENTIAL LEARNING ALGORITHM OF NEURAL NETWORK AND ITS APPLICATION

本文提出了一种序贯学习神经网络,它主要由有界权值调整规则和网络结构自适应调整规则所组成。该网络具有在保持旧知识的前提下有效地序贯学习新输入样本知识的优点。文中给出了这种网络的一种序贯学习算法,详细分析了其学习特性和识别性能。大量的理论分析和实验都证明了网络的有效性。

This paper proposes a sequential learning neural net which consists of the bounded weight adjustment algorithm and structure adaptive adjustment method. This network is characterized by efficiently learning the knowledge of new samples in series. After an efficient sequential learning algorithm of this network, is presented its learning feature and recognition performance in detail are analyzed. The effectiveness of this network have also been shown by theoretical analyses and a lot of experiments in this paper.