A heuristic approach for multiple instance learning by linear separation

Abstract We present a fast heuristic approach for solving binary multiple instance learning (MIL) problem, which consists in discriminating between two kinds of item sets: the sets are called bags and items inside them instances. Assuming that only classes instances allowed, common standard hypothesis states bag is positive if it contains at least negative when all its negative. Our constructs MIL separating hyperplane by preliminary fixing normal reducing phase to univariate nonsmooth optimization can be quickly solved simply exploring kink points. Numerical results presented on set test problems drawn from literature.