一种无师训练的模糊极小极大人工神经网络

提出了一种无师训练的fuzzymin max人工神经网络,它兼有一般fuzzymin max网与ART2网的优点,既弥补了fuzzymin max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,文中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.得到的结论是:对模式识别而言,文中提出的网络比fuzzymin max网和ART2网更具有实用价值.

An unsupervised fuzzy min-max artificial neural network

An unsupervised fuzzy min\|max artificial neural network is proposed in this paper. Our network possesses the strong points, which both the traditional fuzzy min\|max net and the ART2 net have. It not only compensates the weakness which makes the traditional fuzzy min\|max net incapable of learning from any new pattern class, but also overcomes the shortcoming to use too high threshold in ART2. The result of simulating pattern recognition shows that our net can achieve a better effect of pattern recognition with lower threshold than ART2 does with higher threshold. The computational complexity of our net is less than that of ART2. It is concluded that our net is of more practical value than that of fuzzy min\|max net and ART2 net for pattern recognition.