基于自适应邻域选择的正交局部敏感判别分析

维数灾难是机器学习算法在高维数据上学习经常遇到的难题,基于局部敏感判别分析(locality sensitive discriminant analysis,LSDA),可以很好地解决维数灾难问题.且LSDA构建邻域时不能充分反映流形学习对邻域要求和克服测度扭曲问题,利用自适应邻域选择方法来度量邻域,同时,引入施密特正交化获得正交投影矩阵,提出一种自适应邻域选择的正交局部敏感判别分析算法.在ORL和YALE人脸数据库上进行实验,实验结果表明了该算法的有效性.

Orthogonal local sensitive discriminant analysis algorithm based on adaptive neighborhood choosing

The curse of dimensionality is a problem of machine learning algorithm which is often encountered on study of high-dimensional data,while LSDA(locality sensitive discriminant analysis) solve the problem of curse of dimensionality.However,LSDA can not fully reflect the requirements that the manifold learning for neighborhood and overcome the metric distortion problem,by using the adaptive neighborhood selection method to measure the neighborhood,the Gram-Schmidt orthogonalization is introduced to get the orthogonal projection matrix.An adaptive neighborhood choosing of the orthogonal local sensitive discriminant analysis algorithm is proposed.Experimental results verify the effectiveness of the algorithm from the ORL and YALE face database.