cut vm fast semi supervis svm algorithm bin zhao fei wang changshui zhang semi supervis support vector machin vm attempt learn decis boundari travers low data densiti region maxim margin label unlabel exampl tradition vm formul convex integ program problem difficult solv paper propos cut plane semi supervis support vector machin cut vm algorithm solv vm problem specif construct nest sequenc success tighter relax origin vm problem optim problem sequenc effici solv constrain concav convex procedur cccp prove theoret cut vm algorithm take time sn converg guarante accuraci total number sampl dataset averag number featur sparsiti experiment evalu real world dataset show cut vm perform exist vm method effici accuraci 