Hi, It was mentioned in class that support vectors are the points closest to the margin separator. However, when the data set is not easily separable, scaled convex hulls are used to determine the maximum margin separator. But, after we draw the margin separator using points in the reduced convex hull, we see that the points closest to the margin separator were never used anywhere in the computation (because we used scaled hulls). Do they still qualify as support vectors in this case ?

-- Support vector for non-separable datasets

The support vectors will be the scaled convex hull points