We consider the problem of the correspondence leaning of two high-dimensional data sets via the manifold leaning techniques. It's convenient for us to find the shared latent structure of the high-dimensional data sets, if they can be aligned in a uniformed low-dimensional data space. A local affine technique to address this problem is present here. Our method preserves more geometrical knowledge of the high-dimensional data sets and can easily project the new test data into the aligned data space by the local transformation obtained during the training step. The effectiveness of our algorithm ...
