Exactly. Perhaps the paper could have given a clearer message if the abstract had characterized SVMs as a quadratic optimization technique instead of as machine learning?