A Novel Online Learning Algorithm of Support Vector Machines

Support vector machines (SVMs) have been proven as powerful tools in wide variety of learning problems, but it is confronted with the problem of a large amount of computation. In this paper, the existing online learning algorithms of SVMs have been discussed in detail, we analysis the possible changes of support vectors after new samples are added, a novel online learning algorithm of SVMs is presented. The experimental results are given to show that the accuracies of approach is comparable to the batch algorithm, effectively keep classification accuracies, discard useless old samples, and save the memory