Skip to main content
Fig. 8 | Journal of Big Data

Fig. 8

From: Regularized Simple Graph Convolution (SGC) for improved interpretability of large datasets

Fig. 8

The plots showing the results of applying the proposed method of SGC with and without regularization on the features to the dataset of Cora. a Shows the heatmap of the class columns of matrix \(\varvec{\Theta }\) which holds the parameter values for the feature projections of the data \({\mathbf {X}}\) after the inference with SGC without the regularization. b Analogously shows the parameter values but with the inference procedure applying the constraints for the regularization as proposed which produces the shown values of \(\varvec{\Theta }_R\). On the bottom right of each plot there are 7 cells with padded values to produce the heatmaps. The columns of the parameter vector correspond to different classes, each shown separately, and the weights applied to each feature belonging to the nodes. It can be seen that the regularization reduces the amount of weighting over the features highlighting key variables

Back to article page