Abstract:
|
Given a large network and smaller template of interest, we present network matched filters for detection and localization of local structure in the large network that strongly resembles the template. Our proposed matched filters are based on graph matching methodology, and can incorporate the temporal structure present in the time series of networks setting. To wit, to detect and localize template subgraph structure in a large graph, we propose to proceed by treating the template as a small graph to be matched locally to the larger graph. By formulating template detection and localization as a graph matching problem, we can immediately leverage existing robust, scalable graph matching machinery.
|