Non-Parallel Semi-Supervised Classification

A non-parallel semi-supervised algorithm based on kernel spectral clustering is formulated. The prior knowledge about the labels is incorporated into the kernel spectral clustering formulation via adding regularization terms. The proposed method will generate two non-parallel hyperplanes which then are used for the out-of-sample extension. Thanks to the proper decision function, the method can learn the complex structure using a linear kernel and the available labeled data points. The left figure is the result of a purely unsupervised algorithm (KSC) when RBF kernel is used. The figure on the right, shows the result of the proposed semi-supervised method with linear kernel, when the labeled data points are incorporated into the core model (KSC).