Quote:
Originally Posted by mrno5zzz
After learning your kernel method, I have a question. When I do face recognition, there are two different ways. The first one is to extract kernel PCA and then use SVM. The second one is to extract PCA first and then use kernel SVM. Both of them are using kernel method, but what are the differences between them?

In general the two can be viewed as diffrent models. In the former you get a linear combination of (the projection of) some transformedvectors as the weights. In the latter you get a linear combination of the projectedthentransformed vectors as the weights. Hope this helps.