Title: Large Margin Distribution Machine

Abstract: Support vector machine (SVM) has been one of the most popular algorithms in machine learning. The central idea is to maximize the minimum margin, that is, the smallest distance from the instances to the decision boundary. The minimum margin, however, does not necessarily lead to the best generalization in theory, and the margin distribution is proved to be more crucial than the minimum margin for the generalization performance. In this paper, we propose the Margin Distribution Machine (MDM), which tries to get a better performance by optimizing the margin distribution. The margin distribution in MDM is characterized by the first and second-order statistics, i.e., the mean and the variance of the margins. We verify, both theoretically and empirically, the superior performance of MDM.