A Simple Bound for Resilient Submodular Maximization with Curvature (2105.04793v1)

Abstract: Resilient submodular maximization refers to the combinatorial problems studied by Nemhauser and Fisher and asks how to maximize an objective given a number of adversarial removals. For example, one application of this problem is multi-robot sensor planning with adversarial attacks. However, more general applications of submodular maximization are also relevant. Tzoumas et al. obtain near-optimal solutions to this problem by taking advantage of a property called curvature to produce a mechanism which makes certain bait elements interchangeable with other elements of the solution that are produced via typical greedy means. This document demonstrates that -- at least in theory -- applying the method for selection of bait elements to the entire solution can improve that guarantee on solution quality.