Ridge-Adding Approach for SVMpath Singularities

This paper introduces a novel ridge-adding approach for handling the singularity problem which is frequently encountered among the entire regularization path of support vector machine (SVM). Different from the existing ridge-adding method that directly modifies each data point, our approach just adds some random ridge scalars into the primary SVM problem to guarantee the sufficient condition for avoiding singularities. Our proposed method not only provides a simpler implementation but also significantly reduces the influence of the added ridges on the solution path. The experimental results are provided to verify both the efficiency and computational advantages of the proposed method.