Why this algorithm can not recover completely the images that are out of training samples ???
I mean ,if this method can only recover the images in the training examples perfectly,but those new test samples for which there will be a great loss on information even if you've selected all of the eigenvectors??
↧