摘要
概述了目前各种人工神经网络模型的特点及优劣，在此基础上着重介绍一种无导师的在线学习模型——自适应谐振理论ART，并给出其具体算法步骤．详细分析该传统ART算法的实质，指出其合理及不合理处，并就其不合理处提出改进算法（已在AST486上实现）．作为对改进算法的验证，给出了一个“字符识别“的应用实例，改进算法无论在学习过程上还是在识别结果上都比传统算法更类似于人脑．
The good points and the shortcomings of different Artiflcial Neural Network(ANN) models are discussed. and after that, a no-teacher model, Adaptivc Res- onance Theory(ART), is introduced. After the analysis of ART's nature, the drawbacks of ART is pointed out, and a modified algorithm, which is realized in C language in PC-486, is put forward. To illustrate the modified algorithm, an example(alphabet recog- nition) is given, and the result is quite good.
出处
《福州大学学报（自然科学版）》
CAS
CSCD
1993年第5期42-47,共6页
Journal of Fuzhou University(Natural Science Edition)
关键词
人工
神经网络
自适应谐振
自学习
artificial neural networks
no-teacher model
adaptive resonance theory
al. phabet recognition
