【摘要】 提出一种内插法做分类决策(AIMMCD),这种方法使用训练集模式的类标签区分不同模式.与相对于传统的模式识别技术相比,AIMMCD有着自身优势.第一,使用AIMMCD来产生测试模式的类标签时,不需要任何训练过程.即AIMMCD计算效率较高.第二,使用AIMMCD预测真实数据的类标签时,它以合理的方式考虑到训练资料的所有模式中类标签的信息.事实上,该算法假定当训练样本接近一个模式,都会对这个模式类的预测影响很大,而对远离这种模式的训练样本没有什么影响.第三,虽然AIMMCD形式简单,但它不但适用于二分类问题,还可直接用于多分类问题。更多还原

【Abstract】 Pattern recognition techniques have been widely used.This paper proposes an interpolation method for making classification decision(AIMMCD).This method makes an interpolation of the class labels of the patterns of the training set for classifying a new pattern.Compared with conventional pattern recognition techniques,AIMMCD has several advantages.First,when using AIMMCD to produce the class label for the test pattern,not any training procedures are needed.This means that AIMMCD is computationally efficient.Second,when AIMMCD predicts the class label for real-world data,it takes into account the information of the class labels of all the patterns from the training set in a reasonable way.Indeed,the algorithm assumes that the training sample close to a pattern will have much influence on the class prediction of this pattern and the training sample far from this pattern will have little influence.Third,though AIMMCD has a very simple form,it is directly applicable to not only two-class problems but also multi-class problems.更多还原