Tradeoffs on the Efficient Frontier of Network Disruption Attacks

Abstract

A communications network is represented as a graph of flow capacities. We study the problem of finding good network disruption attacks or target sets, i.e., a subset of vertices or edges that, once removed, impede communication between particular nodes. Multiple costs are associated with removing vertices or edges. Success in disrupting communications is traded off against the costs of the attack plans: the efficient frontier of attacks is estimated, and the results are studied in cross-linked diagrams. A multicriterial genetic algorithm is used to discover good plans for disrupting the communications network, where the genes correspond to nodes or links to be attacked. The genetic algorithm is seeded with an initial population of single-target genomes, one for each potential target. Multi-target attacks may be generated by breeding. Being on the efficient frontier guarantees a genome's survival to the next generation, so the population size is allowed to vary. The results are studied in interactive diagrams and in an "aggregate view" of the resulting population. Good attacks were found relatively rapidly, and the aggregate view revealed significant targets