基于判别特征加权的GPLVM算法

Weighted GPLVM Algorithm Based on Discriminant Features

高斯过程隐变量模型是最近提出的比较流行的无监督降维方法。但是,它是一种无监督的机器学习算法,没有突出类结构,使得结果不能有效地表示类别信息。因此,提出一种利用判别特征值对高斯过程隐变量模型进行加权的算法,该算法不仅能够加强模型在低维流形上的判别性,而且能很好地保持类内的流形结构。

Gaussian process latent variable model （GPLVM） is a popular manifold method recently proposed for dimensional reduction. However it cannot keep some class structure of datasets for it is an unsupervised learning method. A weighted GPLVM algorithm will be given using the discriminant features. This algorithm can approve an discrimiant results and keep good manifold in each class.