Beyond Graph Convolutional Network: An Interpretable Regularizer-Centered Optimization Framework

Graph convolutional networks (GCNs) have been attracting widespread attentions due to their encouraging performance and powerful generalizations. However, few work provide a general view interpret various GCNs guide GCNs' designs. In this paper, by revisiting the original GCN, we induce an interpretable regularizer-centerd optimization framework, in which building appropriate regularizers can most GCNs, such as APPNP, JKNet, DAGNN, GNN-LF/HF. Further, under proposed devise dual-regularizer graph network (dubbed tsGCN) capture topological semantic structures from data. Since derived learning rule for tsGCN contains inverse of large matrix thus is time-consuming, leverage Woodbury identity low-rank approximation tricks successfully decrease high computational complexity computing infinite-order convolutions. Extensive experiments on eight public datasets demonstrate that achieves superior against quite state-of-the-art competitors w.r.t. classification tasks.