Research On Fast Support Vector Machines

Support Vector Machines(SVMs)are powerful classification tools,which can be divided into two categories: one is high-efficiency stand-alone SVMs;The other is distributed SVMs.However,the high-efficiency SVMs is still time-consuming.There are still many problems for us to improve the efficiency of SVMs.In the dissertation,the key technique of high-efficiency SVMs are studied.First,in order to solve the time-consuming process of SVMs in stand-alone,we prose Directional Indicator Support Vector Machines(DISVMs)to efficiently identify non-SVs.DISVMs employs a directional indicator,which points to the approximately orthogonal direction of the separating hyperplane,to qualitatively define the location of different samples and thus identify non-SVs.Furthermore,DISVMs leverages a two-stage algorithm:the first stage is to compute the directional indicator.The second stage is to identify nonSVs using the indicator.To avoid misjudgement,we propose CnSV method for non-SVs based on the majority rule.DISVMs screens out non-SVs with light computation and little accuracy loss.Experiments show that our approach significantly reduces the total computation cost.Second,aiming at the shortcomings of poor efficiency of SVMs in distributed environment,we present Multi-Mode Cascade SVMs(MMCascadeSVMs).The parallel arithmetic of DISVMs is proposed to deal with the bottleneck of models in the lowest layer.Moreover,MMCascadeSVMs employs hierarchical similarity,which weights the global model composed by local models,to reduce the useless structure in cascade SVMs.In the training phase,MMCascadeSVMs takes advantage of hierarchical similarity to change the ”reduction tree” structure.In the testing scheme,MMCascadeSVMs employs the early prediction strategy to improve the detection accuracy.Experiments show that our approach is better in terms of training efficiency.