Regularized Locality Preserving Indexing via Spectral Regression中2.2節(jié)采用的SVD克服小樣本問(wèn)題，注意U不是St的非零特征值對(duì)應(yīng)的特征向量，與PCA克服小樣本問(wèn)題不同。
如果該文公式(1)的X have zero mean，則與PCA克服小樣本問(wèn)題(正好降到St的列空間)相同。該文2.1節(jié)也這樣描述
Linear Algebra Methods for Data Mining的lecture5的45頁(yè)PCA和SVD關(guān)系：PCA is SVD done on centered data.