Figure 9

Illustration of the support vector machine method used for regression. Given a set of observed training data (circles and triangles), which are sampled from the hidden original function f(x) (solid line) and may be polluted by noise during this procedure, the SVR constructs the fitted regression function φ(x) (dashed line) by solving the corresponding optimal problem with constrains. The support vectors and non-support vectors are denoted with circles and triangles, respectively.