Current network visualization software relies on the force-directed layout (FDL) algorithm, whose high computational complexity limits visualization of large networks. This paper uses Graph Neural Networks (GNN) to accelerate FDL, achieving a 10 to 100-fold speed improvement and more informative layouts. The speedup is analytically derived and shown to relate to the number of outliers in the adjacency matrix's eigenspectrum, predicting effectiveness for networks with communities and local regularities. A three-dimensional layout of the Internet is generated using GNN, with new measures assessing layout quality and interpretability.

Csaba Both, Nima Dehmamy, Rose Yu, Albert-László Barabási