A stretchy classification methodology adopting multivariate polynomials is proposed in this paper. Through minimization of an approximated p-norm of the parameter vector subject to classification error constraints, an approximated minimum norm solution in dual form is derived for under-determined systems.
This is subsequently transformed into its primal form for over-determined systems. Practical feasibility of the proposed solution is illustrated by an evaluation on synthetic data as well as an application on benchmark real-world data.