Abstract
Recently, Multiple Kernel Learning (MKL) is an interesting research area in kernel machine applications and provides better interpretability and adaptability. Previous works have not considered much about the data itself, especially the intrinsic geometry information of data which is possible being beneficial for machine learning. We propose a manifold regularized multiple kernel machines to use the manifold regularization term to explore the inner geometry distribution of data. In fact, there are some real datasets being embedded in low dimensional manifold being undeveloped or hard to be seen. So adding the manifold regularization term to the original MKL is based on the assumption that the data geometrical distribution information may help to get a proper learning machine performance. We use properties of reproducing kernel Hilbert spaces (RKHS), Representer Theorem and Laplacian Graph method to provide theoretical basis for the algorithm. In experiments, classification accuracies of the algorithm and its ability to represent potential low dimensional manifold are given. Testing results suggest that our proposed method is able to yield competent classification accuracy and worth pursuing further research works.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have