Stochastic Dynamic Games of Asymmetric Information: A Common Information Approach
Abstract
This project aims to develop a systematic theory of stochastic dynamic games of asymmetric information. These games model strategic interactions among autonomous agents operating in uncertain dynamic environments and making decisions based on locally acquired incomplete information. Such scenarios arise in a range of military and security related applications. Military operations involving UAVs or ground vehicles in the presence of adversarial agents, communication of sensitive tactical information in the presence of jamming or eavesdropping agents in a battlefield setting, security of computer and communication networks such as sensing and surveillance networks in the presence of cyber or physical attacks, all involve strategic decision-making based on limited information. Game theory has been widely employed as a modeling and analytical framework for situations where self-interested agents interact in an adversarial, competitive or non-cooperative manner. However, the traditional Nash equilibrium concept ignores the dynamic nature of the game and suffers from issues such as players making non-credible threats or not being sequentially rational. Moreover, with asymmetric information among players, novel effects such as signaling emerge where a player attempts to change the degree of uncertainty of other players by its actions and thus manipulate their behavior to improve its utility. To resolve these unique challenges pre- sented by the dynamics and the information asymmetry in stochastic dynamic games, our goal is to develop a systematic approach for analyzing families of such games. In particular, the objectives of our research are: (i) to identify new sub-classes of Nash equilibria for stochastic games of asym- metric information that capture the sequential, dynamic nature of the game and are amenable to a dynamic program type decomposition for efficient computation, (ii) to find sufficient statistics and structural properties of equilibrium strategies so that they can be efficiently implemented, (iii) to integrate our proposed game-theoretic research with recent results in decentralized control and dynamic teams in order to study non-cooperative interactions among teams of agents, (iv) to apply the concepts and methodologies developed in this research to problems such as control and communication in the presence of adversarial agents, pursuit-evasion games and network security. Central to our approach will be the concept of common information and beliefs on the state of the dynamic system and private information of agents based on the common information. In order to develop our common information approach for stochastic dynamic games, our plan is to proceed as follows: In Year 1, we will focus on games where the common information beliefs are independent of agentsÕ strategies. Based on preliminary results, we expect that our approach can identify novel sufficient statistics and structural results for agentsÕ strategies and provide a dynamic program like backward inductive method for obtaining optimal strategies for all agents. In Year 2, we will focus on more general games where common information beliefs may be strategy- dependent. For these games, we plan to use the common information approach to identify structural properties of equilibrium strategies and construct efficient methods of finding equilibria by using perfect Bayesian equilibria (PBE), sequential equilibria (SE) or public perfect equilibria (PPE) as the solution concept. In Year 3, we plan to combine our approach with decentralized control techniques to investigate team vs. team games and zero-sum games.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 15, 2018
- Source ID
- W911NF1710232
Entities
People
- Ashutosh Nayyar
Organizations
- Army Contracting Command
- Office of the Secretary of Defense
- University of Southern California