Abstract:Classical graph matching aims to find a node correspondence between two unlabeled graphs of known topologies. This problem has a wide range of applications, from matching identities in social networks to identifying similar biological network functions across species. However, when the underlying graphs are unknown, the use of conventional graph matching methods requires inferring the graph topologies first, a process that is highly sensitive to observation errors. In this paper, we tackle the blind graph matching problem with unknown underlying graphs directly using observations of graph signals, which are generated from graph filters applied to graph signal excitations. We propose to construct sample covariance matrices from the observed signals and match the nodes based on the selected sample eigenvectors. Our analysis shows that the blind matching outcome converges to the result obtained with known graph topologies when the signal sampling size is large and the signal noise is small. Numerical results showcase the performance improvement of the proposed algorithm compared to matching two estimated underlying graphs learned from the graph signals.
What problem does this paper attempt to address?
The paper attempts to address the problem of graph matching in the absence of known graph topology (i.e., blind graph matching). Traditional graph matching methods require the topologies of the two graphs to be known. However, in many practical applications, such as social networks, infrastructure networks, and functional brain connectivity, directly observing network links is impossible. Instead, what can be observed are the so-called "graph signals" through the interactions between nodes. These signals can be opinions in social networks, node measurements in infrastructure networks, EEG signals in brain connectivity, or gene network expressions, among others.
When only graph signals are available, a common heuristic for graph matching is to first infer the graph topology from the observed signals and then match the nodes based on the estimated topology. However, this approach is prone to errors because topology inference usually requires strong assumptions about the graph structure or signals. Therefore, this paper proposes a novel blind graph matching method that does not require directly inferring the graph topology or knowing the adjacency matrix or Laplacian matrix in advance.
Specifically, the proposed method assumes that the two sets of graph signals are generated by non-identical graph filters, but these filters exhibit the same low-pass or high-pass spectral trends, meaning that the graph frequency order is preserved. Based on this assumption, the paper proposes matching nodes by computing the eigenbases of the sample covariance matrices of the graph signals and transforming the blind graph matching task into a linear assignment problem, which is then solved using the Hungarian algorithm or a greedy algorithm.
Furthermore, the paper analyzes the performance degradation of blind graph matching relative to the case with known graph topology, quantifying the optimization gap of the objective function and the matching error probability caused by the perturbation of the node similarity matrix due to signal sampling. The study results indicate that with sufficient signal observations and low signal noise, blind graph matching can achieve small matching errors. Additionally, the paper proposes a heuristic method for selecting eigenvectors to reduce the impact of signal sampling perturbations, improve matching accuracy, and significantly reduce computation time.