论文部分内容阅读
提出了一种无师训练的ｆｕｚｚｙ ｍ ｉｎｍ ａｘ 人工神经网络，它兼有一般ｆｕｚｚｙ ｍ ｉｎｍ ａｘ 网与ＡＲＴ２网的优点，既弥补了ｆｕｚｚｙ ｍ ｉｎｍ ａｘ 网不能自适应在线学习新类的缺陷，又消除了ＡＲＴ２网警戒门限过高的弊病。经模式识别仿真对比，对同样的输入数据，我们提出的网络用较低的警戒门限值即可达到ＡＲＴ２用很高的警戒门限值才能达到的分类效果，且计算量大大减少。对模式识别而言，所提出的网络比ｆｕｚｚｙ ｍ ｉｎｍ ａｘ 网和ＡＲＴ２网更具有实用价值。
A fuzzy m in-m ax artificial neural network without teacher training is proposed, which combines the advantages of the general fuzzy m in-m ax network with the ART2 network, which not only makes up the fuzzy m in m ax network can not adapt to online learning The new class of defects, but also eliminates the shortcomings of ART2 network warning threshold too high. Compared with the pattern recognition simulation, the proposed network can achieve the classification effect that can be achieved by using the high warning threshold value with the lower warning threshold for the same input data, and the calculation amount is greatly reduced. For pattern recognition, the proposed network is more practical than fuzzy m in-m ax network and ART2 network.