该算法是无监督的，贡献很大。很难得解决了最大边缘的不需要依赖标签信息，就能获得最大边缘的超平面映射-projecting data samples into the most discriminative subspace, where clusters are most well-separated。