Online Nearest Point Algorithm for L2-SVM

During last few years, a number of kernel-based online algorithms have been developed that have shown better performance on a number of tasks. A well designed online algorithm needs less computation to reach the same test accuracy as the corresponding batch algorithm. In this paper, we devise an online training algorithm for L2-SVM. Our work is motivated by HULLER, an online algorithm proposed by A. Bordes and L. Bottou. The proposed algorithm implements two speedups with respect to HULLER, first it chooses an old example for removal based on sound computation instead of random selection; second it uses more effective update rule. Experiments on benchmark data sets show the merits of our method.