What is the representer theorem in SVM?

The representer theorem states that SVM solutions can be expressed as a combination of training samples.

It ensures that the optimal decision function lies in the span of training data, meaning weights depend only on support vectors. This is why kernel methods work efficiently in SVM.