对有多个来源的数据集进行充分降维, 文献中常见的方法是利用分类变量信息并融入先验知识或者鉴于混合模型分别估计不同成分的中心子空间. 文章主要借鉴了普通线性模型的 Maximin估计思想, 提出了中心子空间的 Maximin 方向估计, 以减少数据来源 较多而呈现的复杂性. 模拟结果显示, Maximin方向估计能够有效地探索子总体的共性.

To explore the different traits of heterogeneous data from multiple sources in sufficient dimension reduction, one common method in the literature is based on categorical predictors and prior knowledge, the other is estimating the central subspace of different components separately in mixture models. In view of the maximin estimator for Gaussian linear models, this paper proposes the maximin direction estimator, which reduces the complexity from various data sources. Our simulation results show that the proposed method can detect the homogeneity from the subgroups in a more efficient way.