CAREER: Evolutionary Games in Dynamic and Networked Environments for Modeling and Controlling Large-Scale Multi-agent Systems

点击翻译按钮获取中文摘要

Classical game theory addresses how individuals make decisions given suitable incentives, for example, whether to use resources rapaciously or with restraint. However, game theory does not typically address the consequences of the actions that reshape the resources over the long term. Indeed, individuals' actions often subsequently modify the commons (environment) and associated payoffs. In this project, we propose a unified mathematical framework to model and analyze the coupled evolution of individuals' incentives, opinions, and the environment using tools from game theory, network science, and nonlinear dynamic systems. Based on the mathematical framework, the proposed project is organized to study fundamental issues relating to (a) when and how desirable behavior, e.g., cooperative behavior, arise in the populations, and (b) whether tragedies of the commons can be averted in complex systems, e.g., during a pandemic. Scientific contributions of this project will have the potential to have a transformational impact on our understanding of the emergence of cooperation and environmental collapse in public health systems where individuals' actions affect the resources, and in engineered multi-agent systems, e.g., autonomous or energy systems, that involve self-interested entities. The overarching goals of the project are rooted in an educational agenda with initiatives, e.g., a summer residential research experience for educators, designed to expose the broader public to central concepts in game theory and nonlinear systems, and push for a systems-thinking perspective on societal problems.The premise of this project is that individual behavior is dynamic, i.e., evolves according to selection or learning, and such learning behavior has subsequent effects on the environment, and thus on the future incentives for learning. The proposed research is a concerted effort to develop a mathematical framework for studying population behavior when the population’s well-being depends on the environment that the behavior is affecting. The proposed research aims to achieve the following scientific contributions: 1) novel models of strategic learning dynamics in feedback-evolving games with relevance to socio-biological and -technological systems including epidemics and autonomous systems; 2) decentralized algorithms for tracking rational behavior in dynamic network games; 3) a framework for dynamic intervention mechanisms to induce desirable system-level behavior in such settings; 4) design and analysis of experiments to uncover the role of peer effects and ambiguity on perceived risks on cooperation. This effort will lead to novel analysis, and scalable decentralized algorithms applicable to addressing real-world problems in social and technological multi-agent systems.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.