Research

I am broadly interested in the analysis and control of multi-agent systems, where self-interested individuals make decisions in an environment of other decision-makers. A major theme in my research concerns identifying the role of information in these complex environments. Below are a few areas of my research.

Strategic decision-making in adversarial environments

A system's reliance on multiple, interconnected sub-components becomes a liability in the presence of adversarial threats. Effective security implementations are an increasingly important aspect in system design. Ensuring security comes with a host of challenges. For one, the amount of available resources to devote to security is limited, and hence strategic allocation becomes an important consideration. Moreover, the strategic behavior of an adversary may be unknown due to informational limitations. Thus, security measures must be effective even under uncertainty.

My research has developed ``Colonel Blotto games" as a unifying framework to study various elements in competitive resource allocation problems. Colonel Blotto games are a class of zero-sum games that naturally illustrate these problems, and are suitable for many applications, e.g. political campaigns, market competitions, and military conflicts. The Blotto game has been studied since game theory first emerged as a discipline (Borel introduced the first formulation in 1921), yet general equilibrium solutions have been derived only in recent years.

A focus of our work is the analysis of informational asymmetries between competitors, and characterizing the "value of information", e.g. the performance improvement that is attainable from having some unit of information. These scenarios also factor into design implementations. For example, how should a pool of assets be delegated to multiple teams tasked with accomplishing different objectives?

Influence of networked multi-agent systems

The interactions of multiple, independent decision-makers forms the core of a networked multi-agent system. While direct control of these systems, such as social networks, smart grids, and transportation networks is not possible, influence mechanisms can be applied at the individual level. This influence can be magnified by cascading effects through a network.

My research has focused on identifying influence policies, e.g. algorithms, that can be applied to a networked system such that its emergent behavior is desirable. Our work has identified trade-offs in the design of local decision-making rules that can mitigate the effects of adversarial influences, but subjects the system to inefficiencies. The role of information is also central to this analysis. How does an adversary's knowledge (or lack thereof) about the network impact its ability to spread its influence through the network?

Social behavior in epidemics

Infectious diseases pose a major risk to global health. They spread through physical contact between infected and susceptible individuals. These interactions can be modeled by a network whose edges describe the physical connections. Behavior also plays a role in how an epidemic runs its course. Individuals that receive information about the disease from social contacts or broadcasts can take preventative measures. Our work has analyzed dynamical models of epidemic spread over networks where the agents, or nodes in the network, are informed of disease prevalence through social contacts that may or may not coincide with their physical contacts. Based on their information, the agents take preventative measures by social distancing - reducing contact with their physical neighbors to lower the probability of getting infected.

Evolutionary game theory

Evolutionary game theory studies the proliferation and extinction of interacting species in a rigorous mathematical framework. One of the central ideas is the evolution of cooperation - how individuals come to work together to provide or improve a common good.

Our work has investigated feedback mechanisms between incentives for action and the environment. That is, individual actions can affect environmental conditions, and in turn the changing environment shapes incentives for future actions. We find such a co-evolutionary process provides rich and complex dynamics, whose outcomes include environmental depletion or sustainability. Our insights provide a new avenue towards the control of sustainable human-resource systems.

Organisms whose survival rely on coordinated efforts often have noisy sensing of the environment as well as of the perceptions of their neighbors. Communication systems can serve as a coordination device. For example, some cooperative behaviors in bacteria are facilitated by quorum sensing. Our work has provided a principled game-theoretic approach to evaluating the role of information sharing in group decision-making under uncertain conditions.