Fortification Against Cascade Propagation Under Uncertainty

Network cascades represent a number of real-life applications: social influence, electrical grid failures, viral spread, and so on. The commonality between these phenomena is that they begin from a set of seed nodes and spread to other regions of the network. We consider a variant of a critical node detection problem dubbed the robust critical node fortification problem, wherein the decision maker wishes to fortify nodes (within a budget) to limit the spread of cascading behavior under uncertain conditions. In particular, the arc weights-how much influence one node has on another in the cascade process-are uncertain but are known to lie in some range bounded by a worst-case budget uncertainty. This problem is shown to be NP-hard even in the deterministic case. We formulate a mixed-integer program (MIP) to solve the deterministic problem and improve its continuous relaxation via nonlinear constraints and convexification. The robust problem is computationally more difficult, and we present an MIP-based expand-and-cut exact solution algorithm, in which the expansion is enhanced by cutting planes, which are themselves tied to the expansion process. Insights from these exact solutions motivate two novel (interrelated) centrality measures, and a centrality-based heuristic that obtains high-quality solutions within a few seconds. Finally, extensive computational results are given to validate our theoretical developments as well as provide insights into structural properties of the robust problem and its solution.