This project develops representations and techniques for analysis of complex interactions among groups of different types (e.g., social, financial, geo-political, military, criminal) with potentially overlapping membership and conflicting interests, and some adversarial in nature.
The groups may (strongly or weakly) interact on a community level while the members of the same and different groups may (strongly or weakly) interact on an individual level. The scenarios of interest are inherently compositional and multi-scale, and we seek a game-theoretic characterization of strategic network behavior.
We build on several core methods to address (1) inference problems: identification of multi-scale network structure from potentially incomplete observational data, as well as (2) decision/control problems: design of effective control and intervention schemes at appropriate levels of the network in order to induce desirable individual as well as group behaviors.
Methods to be employed
We pursue three tightly coupled research thrusts.
(T1) Multi-scale Game Analysis lays the theoretical foundation by developing a general analytical framework for multi-scale games, through iterative abstraction and refinement techniques, and by exploiting limited role symmetry and multi-scale empirical game-theoretic analysis.
(T2) Network and Community Inference focuses on developing methods to extract multi-scale network structures from partial observational data, especially in adversarial scenarios; we will develop methods for estimating hidden network attributes, including links, node classes, and community structure.
(T3) Multi-scale Decision and Intervention investigates how the unique multi-scale structure of the underlying network may enable multi-scale control and intervention mechanisms using techniques developed under T1 and T2; we will examine the stability of the equilibria in multi-scale network games, and design control schemes and intervention strategies applied at different scales that involve the removal or weakening/strengthening of links/nodes as well as entire communities to attain desirable equilibria.
The significance of the proposed effort to the advancement of knowledge
The proposed multi-scale framework exploits and extends existing representations of graphical games. The multi-scale model is accompanied by a multi-scale analysis and computational framework. Anticipated outcomes include formal methodologies, algorithms, performance bounds, and algorithmic implementations. They will be reported in journal and conference publications; all software developed will be made available to the research and DoD community.
We will advance the basic science for a new compositional game theory and transition the output to practice. Furthermore, we will train skilled human resources.
Our proposed research will also substantially contribute to the agility of DoD responses to a broad spectrum of real-world security risks. Risk heterogeneity is fundamentally tied to scale; at two extremes, it spans nation states and lone-wolf actors. By explicitly performing analysis and decision making at multiple scales, we enable the conceptualization and development of agile solutions.