CensNet: Convolution with Edge-Node Switching in Graph Neural Networks

In this paper, we present CensNet, Convolution with Edge-Node Switching graph neural network, for semi-supervised classification and regression in graph-structured data with both node and edge features. CensNet is a general graph embedding framework, which embeds both nodes and edges to a latent feature space. By using line graph of the original undirected graph, the role of nodes and edges are switched, and two novel graph convolution operations are proposed for feature propagation. Experimental results on real-world academic citation networks and quantum chemistry graphs show that our approach has achieved or matched the state-of-the-art performance.

Switching in One-Factorisations of Complete Graphs

We define two types of switchings between one-factorisations of complete graphs, called factor-switching and vertex-switching. For each switching operation and for each $n\le 12$, we build a switching graph that records the transformations between isomorphism classes of one-factorisations of $K_{n}$. We establish various parameters of our switching graphs, including order, size, degree sequence, clique number and the radius of each component.As well as computing data for $n\le12$, we demonstrate several properties that hold for one-factorisations of $K_{n}$ for general $n$. We show that such factorisations have a parity which is not changed by factor-switching, and this leads to disconnected switching graphs. We also characterise the isolated vertices that arise from an absence of switchings. For factor-switching the isolated vertices are perfect one-factorisations, while for vertex-switching the isolated vertices are closely related to atomic Latin squares.