Abstract: The algorithm for solving large-scale support vector machine(SVM) needs large memory requirement and computation time.Therefore,large-scale SVMs are performed on computer clusters or supercomputers.An efficient algorithm for large-scale SVM was presented,which can be operated on a daily-life PC.First,the large-scale training examples were subsampled to reduce the data size.Then,the random Fourier mapping was explicitly applied to the subsample to generate the random feature space,making it possible to apply a linear SVM to uniformly approximate to the Gaussian kernel SVM.Finally,a parallelized linear SVM algorithm was implemented to speed up the training further.Experimental results on benchmark datasets demonstrate the feasibility and efficiency of the proposed algorithm.

Key words: Large-scale support vector machine,Subsampling,Random Fourier features,Parallelized linear SVM