The Support Ve ct or Machi ne So fa r we have used a reference as sumpt ion tha t there exists a linear classifier that has a larg e ge ometric margin, i.e., whose decision b oundar y is we ll separa ted from all the training images (e xampl