Using Sensitivity Analysis for Designing Resilient Systems

Abstract

In this paper, we present the research results on sensitivity analysis in probabilistic graphical models, in particular Bayesian networks. We look at both the theory and application of sensitivity analysis in different domains, such as modeling (to represent the different properties and relationships between the variables), inference (to provide exact or approximate answers to user queries), and analysis (to summarize current behavior, predict future trends, and suggest actions for achieving certain targets). We also discuss using sensitivity analysis for the problem of designing systems that are resilient, such that the systems are resistant from large-scale perturbations caused by unexpected events and changes, and if their functionality is lost temporarily due to outside forces, the systems can recover gracefully and quickly to restore their functionality in the long run.