Robust Multiple Instance Learning