Abstract: This paper defines a different distance measurement between points of inner class and points of different ones to replace the Enclidean distance measurement through introducing an supervised learning strategy. This paper realizes an improved Laplacian Eigenmap algorithm and puts the lower dimension results as the input of BP neural network to realize classification. Experimental results indicate that the improved Laplacian Eigenmap algorithm reduces the training time of BP neural network and has a better classification result.

Key words: Laplacian Eigenmap, supervised learning, classifier, dissimilarity

摘要： 引用监督学习策略，定义类内和类间不同的距离度量方式，以替代原来的欧式距离度量，实现对拉普拉斯特征映射算法的改进。将降维之后的结果作为BP神经网络的输入，实现分类。实验结果表明，基于改进的拉普拉斯特征映射算法降维之后的结果，减少了神经网络的训练时间，具有较好的分类正确率。

关键词: 拉普拉斯特征映射, 监督学习, 分类器, 相异度

CLC Number: 

• TP391