Final Report: Data-Driven Game Theory

Abstract

We make multiple contributions in this research. Though classical game theory assumes that payoff matrices are given for the players involved in the game, we show how these can be learned automatically from a body of data. We further show new definitions of quasi-polynomially computable approximate equilibria and show how to efficiently and approximately compute them. We apply these methods to the behavior of real world terrorist groups. Continuing with our work on adversarial models, we study methods to disclose information publicly in order to shape the adversary's actions to our advantage. We further develop methods to quantify and reduce lethality of networks. Our work develops and further studies the use of game-theoretic frameworks to the study of security games, online games, as well as games involving diverse terrorist groups including Lashkar-e-Taiba and the Indian Mujahideen.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Aug 15, 2016
Accession Number
AD1055617

Entities

People

  • V. S. Subrahmanian

Organizations

  • University of Maryland

Tags

Communities of Interest

  • Autonomy

DTIC Thesaurus Topics

  • Agreements
  • Algorithms
  • Artificial Intelligence
  • Bayesian Networks
  • Computational Science
  • Computers
  • Counterterrorism
  • Data Sets
  • Game Theory
  • Intelligent Agents
  • Intelligent Systems
  • Linear Programming
  • Machine Learning
  • Multiagent Systems
  • Probability
  • Students
  • Supervised Machine Learning

Fields of Study

  • Computer science

Readers

  • Computer Vision.
  • Game Theory.
  • Political Violence and Terrorism Studies.