无师训练Fuzzy Min-Max人工神经网络

提出了一种无师训练的ｆｕｚｚｙ ｍ ｉｎｍ ａｘ 人工神经网络，它兼有一般ｆｕｚｚｙ ｍ ｉｎｍ ａｘ 网与ＡＲＴ２网的优点，既弥补了ｆｕｚｚｙ ｍ ｉｎｍ 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 ｉｎｍ ａｘ 网和ＡＲＴ２网更具有实用价值。

A Unsupervised Fuzzy Min-Max Artificial Neural Network

A unsupervised fuzzy m in m ax artificialneuralnetwork is proposed in this paper. Our network pos sesses the strong points both of the traditionalfuzzy m in m ax net and the ART2 net. It not only counteracts the weakness which m akes the traditionalfuzzy m in m ax netto be incapable oflearning from any new pattern class, but also overcom es the shortcom ing w hich causes using too high vigilance threshold in ART2. The resultofsim ulating pattern recognition show sthatournetcan achieve abettereffectofpattern recognition w ith lowerthreshold than that ART2 does w ith higherthreshold using the sam e data. The com putationalcom plexity ofournetislowerthan thatof ART2. Ourconclusion isthatournetpossessesm ore practicalvaluethan thatoffuzzy m in m ax netand ART2 netfor pattern recognition.