Abstract.ThetutorialstartswithanoverviewoftheconceptsofVCdimensionandstructuralriskminimization.WethendescribelinearSupportVectorMachines(SVMs)forseparableandnon-separabledata,workingthroughanon-trivialexampleindetail.Wedescribeamechanicalanalogy,anddiscusswhenSVMsolutionsareuniqueand