A multi-instance learning algorithm based on nonparallel classifier

In this paper, we proposed a new Multiple-Instance Learning (MIL) method based on nonparallel classifier (called MI-NSVM). The method is mainly divided into two steps. The first step is to generate a spare hyperplane and estimate the score of each instance in positive bags. For the second step, MI-NSVM seeks the “most positive” instance of each positive bag by the information obtained in the first step, and then generates the second hyperplane. MI-NSVM is a useful extension of twin SVM and has the same advantages as it. All experiments show that our method is superior to the traditional MI-SVM and MI-TSVM in both computation time and classification accuracy.