Incremental and Decremental LS-SVM for Function Estimation

摘要

This article proposes an incremental and decremental LS-SVM (Least Square Support Vector Machines) for function estimation, which can make use of the current information plus the new data in an incremental formation, and prune the LS-SVM to get sparse approximation online. When a SV is added or removed, the incremental and decremental formation avoids large-scale matrix inversion operation. Thus the computation cost is reduced and the online training becomes possible. The experiments of function estimation on the artificial and real data have shown the feasibility.