Elizabeth Andreas (Dept. of Mathematical Sciences, MSU)

4/11/2024  3:10pm

Abstract: 

An adversary decision process can be represented as a multilevel network, where each node of the network has multiple states an adversary can be in. This project aims to utilize a previously developed framework by Gan et al. [1] that creates a virtual node for each node state and describes the virtual nodes using a quasi-Boolean function representing how one adversarial state leads to another. Then a graph coined the expanded network can be used to find fixed point and complex attractors of the system. Additionally, due to the structure of the expanded network, we show tokening methods traditionally used in Petri nets can be used to analyze adversarial decision flows given different initial states of the decision process. In this project, we also use perturbations of the quasi-Boolean functions to analyze the neighbors of an expanded network, giving insight into the adversaries decision-making process given a slightly different representation of the adversarial decision-making process. Thus, we can analyze adversarial dynamics without requiring the assumption that we know the adversaries decision-making process perfectly and instead allow for analysis using an initial “best guess” approach. The complexity of the adversarial decision-making process can be measured by studying how the attractors and tokening results change as we look at neighbors.