PhD Defense: Computational Analysis of Intelligent Agents: Social and Strategic Settings

The goal of this study is to analyse and predict the behaviour of multiple interacting intelligent agents. We propose and analyze theoretical models for behavior of such agents in social and strategic settings.
In social networks, we generalize and get new theoretical results for scheduling a cascade in a network and for seed set selection for influence maximization. From the relatively simple models of influence maximization, we move on to a much richer model -- that of strategic decision making in hypergraphs. In this setting, the agents have a cost associated with resource allocation along hyperedges. In return, the agents derive a reward from the hyperedges. The interaction among the agents is implicitly specified by these cost and reward functions. We give one of the first positive results on computability of equilibrium in this setting. Efficient computability is an important prerequisite for solution concepts. Therefore, it is an important first step in analysis of such models. We give a convex formulation for the problem. This opens the possibility that that various dynamics, e.g. best-response, may converge to equilibrium. This is a direction of further research. We also contribute the first combinatorial algorithm for computing an equilibrium for the classical Cournot Oligopoly. In strategic games, we study and develop new solution concept for the case when there is considerable uncertainties in payoffs. In addition, we apply our framework for analysis of a real-world strategic game involving five players.
Game theory for strategic games makes many assumptions. Specifically, it almost always assumes that payoffs are given. However, with increased availability of data, it should now be possible to mine human incentives from data. To achieve this we develop a new data analytic framework for finding payoffs for a game.
Examining Committee:
Chair: Dr. V.S. Subrahmanian
Co-Chair: Dr. Mohammad Hajiaghayi
Dean’s rep: Dr. Subramanian Raghavan
Members: Dr. Sarit Kraus
Dr. Dana Nau