In this paper we use the Birkhoff\char21{}von Neumann decomposition of the diffusion kernel to compute a polytopal measure of graph complexity. We decompose the diffusion kernel into a series of weighted Birkhoff combinations and compute the entropy associated with the weighting proportions (polytopal complexity). The maximum entropy Birkhoff combination can be expressed in terms of matrix permanents. This allows us to introduce a phase-transition principle that links our definition of polytopal complexity to the heat flowing through the network at a given diffusion time. The result is an efficiently computed complexity measure, which we refer to as flow complexity. Moreover, the flow complexity measure allows us to analyze graphs and networks in terms of the thermodynamic depth. We compare our method with three alternative methods described in the literature (Estrada's heterogeneity index, the Laplacian energy, and the von Neumann entropy). Our study is based on 217 protein-protein interaction (PPI) networks including histidine kinases from several species of bacteria. We find a correlation between structural complexity and phylogeny (more evolved species have statistically more complex PPIs). Although our methods outperform the alternatives, we find similarities with Estrada's heterogeneity index in terms of network size independence and predictive power.
22 February 2008 in Pattern Analysis and Applications
In the ideal (homophilic) regime of vanilla GNNs, nodes belonging to the same community have the same label: most of the nodes are harmonic (their unknown labels result from averaging those of their neighbors given some labeled nodes). In other words, heterophily (when neighboring nodes have different labels) can be seen as a "loss of harmonicity". In this paper, we define "structural heterophily" in terms of the ratio between the harmonicity of the network (Laplacian Dirichlet energy) and the harmonicity of its homophilic version (the so-called "ground" energy). This new measure inspires a novel GNN model (Diffusion-Jump GNN) that bypasses structural heterophily by "jumping" through the network in order to relate distant homologs. However, instead of using hops as standard High-Order (HO) GNNs (MixHop) do, our jumps are rooted in a structural well-known metric: the diffusion distance. Computing the "diffusion matrix" (DM) is the core of this method. Our main contribution is that we learn both the diffusion distances and the "structural filters" derived from them. Since diffusion distances have a spectral interpretation, we learn orthogonal approximations of the Laplacian eigenvectors while the prediction loss is minimized. This leads to an interplay between a Dirichlet loss, which captures low-frequency content, and a prediction loss which refines that content leading to empirical eigenfunctions. Finally, our experimental results show that we are very competitive with the State-Of-the-Art (SOTA) both in homophilic and heterophilic datasets, even in large graphs.