Stopping Criteria for SVMs

The support vector machine has been shown to be comparable to the state of the art for classification. In certain situations however, the convergence to a global optimum of the optimization problem can be very slow. It is shown that solutions suboptimal in terms of the optimization problem can still have very good performance. By using a heuristic stopping criterion (termed HERMES) these solutions can be found in substantially less time than is required to attain a global optimum of the optimization problem.