RI:Small:Collaborative Research:Influence Games: A Game-theoretic Approach to Strategic Behavior in Networks

点击翻译按钮获取中文摘要

The world is becoming increasingly interconnected. While not all connections have the same level of importance or even the same meaning, these connections nevertheless play a crucial role in one's behavioral and lifestyle choices. This is particularly striking in strategic scenarios where individual choices are interdependent on each other. This research seeks to model how networked individuals influence each other in their decision making and what collective outcomes may arise from such a system of influence. It seeks to advance our scientific knowledge of strategic behavior in networks by making the models more realistic, allowing for changes over time, and considering the underlying context. These advances are important in part because of their potential impact in a wide range of domains including public health policy, smart power grid, and financial systems. In addition, the project will contribute to the educational enrichment of undergraduate students, including underrepresented and first-generation students. It will bring together two distinct groups of students, namely undergraduate liberal arts students and graduate computer science students, under a symbiotic collaboration plan. The research results will be broadly disseminated through a website and will also be integrated into undergraduate- and graduate-level education.This project investigates several important open directions in the computational game-theoretic study of influence in networks. It will address a variety of fundamental research problems, including the challenge of identifying "most influential" individuals in a network. In particular, the research has three major parts: (1) The challenge of complexity: design game-theoretic models of influence in networks to allow (a) flexibility in behavioral choices (from multiple, non-binary discrete choices to a continuum of behavioral choices) and (b) non-linear influences without any restriction on polarities (positive/negative). (2) The reality of dynamics: model dynamic evolution of influence networks. (3) The power of context: model the contextual environment of strategic behavior. In these three thrusts, the project significantly departs from the well-studied approaches to influence maximization as well as the traditional centrality measures in social network analysis. It seeks to design network-aware algorithms, including provable approximation algorithms and practical heuristics, for computing stable outcomes and identifying most influential individuals in a network relative to a desirable outcome. Ultimately, the research seeks to provide computational tools for policy analysts to perform minimal targeted interventions in a social network for achieving a desirable social outcome. To that end, the project will use real-world behavioral data. It will employ, adapt, or extend existing machine learning algorithms to learn context-aware models without imposing any restriction on the structure of the networks.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.