摘要
从输入的原始信息得到特征通常需要复杂的非线性运算,直接找到这种算法是很困难的。而M-P神经元模型的几何意义指出:构造一个网络,使对给定的样本集能进行符合要求的分类,等价于求出一组领域,对给定样本集中的点,能按分类的要求用领域覆盖将它们分隔开来。但是,在实际的大规模应用中,如时间序列预测的典型问题———股票预测,其给定的样本集中可能含有异动点,会引起错误的学习结果,因此,有必要引入自组织和概率决策化方法,提高分类的正确性,同时还可降低神经网络的结构规模,提高识别的速度。作者给出一种构造性的概率决策自组织分类器SPDC(ASelf-adjustingandProbabilisticDecision-makingClassifier),重点讨论了在覆盖算法中引入自组织和概率决策化方法。
Usually getting the characteristic needs the complicated nonlinear operation from the raw information imported.It is very difficult to find directly this kind of algorithm. And the geometry meaning of M-P's neuron model is pointed out: that to classify sample according to the requirement by to construct neural networks is equal to find of one team of domain, and the points of the preset sample are separated with the covering domain. But to some applications, such as the typical problem of time queue forecasting - stock share forecasting, because its preset sample set are concentrated probably to contain some exceptions and cause wrong results. Therefore it is necessity to introduce the self-adjusting and probabilistic decision-making for raising correct rate of classification, at the same time the structure can degrade neural network scale and raise the speed recognized. A classifier of self-adjusting and probabilistic decision-making is supplied in this paper. It is discussed importantly to introduce the self-adjusting and probabilistic decision-making in the covering algorithm.
出处
《微机发展》
2003年第7期85-87,90,共4页
Microcomputer Development
基金
国家自然科学基金资助项目(60175018)
