Abstract by Charles Johnson
Source-Centered Linear Dynamical Network Modeling
We discuss a class of linear dynamical networks that can be represented by dynamical structure functions (DSFs). We construct notions of distance and effect for such linear dynamical networks. We consider the influence of one source node on this network and use our notion of distance to construct a network representation which simplifies the calculation of the source node’s influence in the network. This new representation is directed and acyclic and its structure has levels of distance from the chosen source node. Additionally, its boolean structure is a substructure to that of the DSF representation of the network. We show how to learn this network representation from data. We demonstrate that learning this structure is an easier problem than reconstructing the DSF of the system.