Abstract
Most existing semi-supervised methods implemented either the cluster assumption or the manifold assumption. The performance will degrade if the assumption was not proper for the data. A method was proposed by combining both the cluster assumption and the manifold assumption. A semi-supervised kernel which reflected geometric information of the samples was constructed through warping the Reproducing Kernel Hilbert Space. Then the semi-supervised kernel was used in SVM which was based on cluster assumption, and a progressive learning procedure was used in the proposed method. Experiments had been took on synthetic and real data sets, and the results showed that, compared with the progressive SVM with common kernel and the standard SVM with semi supervised kernel, the proposed method using semi-supervised kernel in progressive SVM had competitive performance.