Fig. 3
From: Regularized Simple Graph Convolution (SGC) for improved interpretability of large datasets
The above plots show the results of applying the SGC method with and without regularization upon the synthetic circular data. In a the dashed line shows the separation line where each side defines the point labels and the solid lines shows the SGC projections learned in \(\varvec{\Theta }\), and perfect accuracy is achieved. b Presents multiple subplots of separate independent runs of the proposed regularized SGC methodology where there are different random initializations using gradient descent. Various comparable fittings are found and it can be seen how all the aspects of the regularization upon \(\varvec{\Theta }_R\) are respected in terms of the magnitude, relative directions between class vectors and the number of components (features) used