Budgeted semi-supervised support vector machine

Due to the prevalence of unlabeled data, semi-supervised learning has drawn significant attention and has been found applicable in many real-world applications. In this paper, we present the so-called Budgeted Semi-supervised Sup-port Vector Machine (BS3VM), a method that leverages the excellent generalization capacity of kernel-based method with the adjacent and distributive information carried in a spectral graph for semi-supervised learning purpose. The fact that the optimization problem of BS3VM can be solved directly in the primal form makes it fast and efficient in memory usage. We validate the proposed method on several benchmark datasets to demonstrate its accuracy and efficiency. The experimental results show that BS3VM can scale up efficiently to the large-scale datasets where it yields a comparable classification accuracy while simultaneously achieving a significant computational speed-up compared with the baselines.