Graph Diffusion & PCA Framework for Semi-supervised Learning

A novel framework called Graph diffusion & PCA (GDPCA) is proposed in the context of semi-supervised learning on graph structured data. It combines a modified Principal Component Analysis with classical supervised loss and Laplacian regularization, thus handling case where adjacency matrix Sparse avoiding Curse dimensionality. Our can be applied to non-graph datasets as well, such images by constructing similarity graph. GDPCA improves node classification enriching local structure covariance. We demonstrate performance experiments citation networks images, we show that compares favourably best state-of-the-art algorithms has significantly lower computational complexity.