Topics in Learning and Information Dynamics in Game Theory
Open Access
Author:
Young, Matthew
Graduate Program:
Mathematics
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
June 19, 2020
Committee Members:
Andrew Leonard Belmonte, Dissertation Advisor/Co-Advisor Andrew Leonard Belmonte, Committee Chair/Co-Chair Christopher H Griffin, Committee Member Jan S Reimann, Committee Member Syed Nageeb Ali, Outside Member Sergei Tabachnikov, Committee Member Alexei Novikov, Program Head/Chair
Keywords:
Game Theory Learning Agent Based Models Neural Networks Game Theory Learning Agent Based Models Neural Networks
Abstract:
We discuss the role of learning and information in game theory, and design and investigate four models using different learning mechanisms. The first consists of rational players, one of which has the option of purchasing information about the other player's strategy before choosing their own. The second is an agent-based public goods model where players incrementally adjust their contribution levels after each game they play. The third is an agent-based rock, paper, scissors model where players choose their strategies by flipping cards with a win-stay lose-shift strategy. The fourth is a machine learning model where we train adversarial neural networks to play arbitrary 2 x 2 games together. We study various aspects of each of these models and how the different learning dynamics in them influence their behavior.
