Our world is comprised of many different large-scale systems, in which autonomous agents interact with one another in complex ways, many times without any centralized direction. Only recently have we begun a rigorous study of these systems in order to leverage their robustness, efficiency, and local simplicity to apply to our own engineered systems. For instance, the rapidly growing internet of things is a sprawling network of interconnected devices, some automated and some manually-controlled, each with its own computational abilities, data sets, and objectives. Another example is traffic systems of the near-future, where many vehicles (human- and computer-driven) must interact with one another, each having its own destination and acting selfishly, based on incomplete information. Whatever the application, the designer of such systems faces the following challenges:

1. Each agent is equipped with its own local objective, local computational resources, local data, and local time scale.

2. Even small and simple adjustments to local behavior can have cascading and undesirable effects on the entire system.

3. The interconnected nature of the system leaves it open to new vulnerabilities that can be exploited by an attacker

An understanding of the design and behavior of these large-scale systems is imperative. My research blends ideas from computer science, engineering, and economics to create new and exciting theory for large-scale autonomous systems. In addition, my focus is to apply this theory to novel domains in order to benefit society.

Understanding the Value of Information in Distributed Architecture

A system designer must balance tradeoffs between several competing objectives: for instance, allowing agents to share information leads to a higher system performance, but is more costly to build and maintain. Understanding the value of information sharing and what distributed architectures ensure high performance is key to design strategy.

Designing Resilient and Secured Network Systems

Agents in the system are interdependent: the decision of one depends on the state of others. While these distributed architectures are generally robust against a single point of failure, they also introduce new vulnerabilities. An attack applied to one decision-maker can have cascading effects to the system. Likewise, a small but strategic attack against many key decision-makers can stealthily drive the system to an undesirable state. The vulnerabilities that exist and architectures that lead to lower risk are therefore important topics of study.

Teaming Human and Software

Many large-scale systems require heterogeneous decision-makers to interact with one another, sometimes including humans and computers. While a system designer may have the capability to specify the behavior of some decision-makers, she may only be able to incentivize the behavior of others. What information ought to be shared between software and humans and how to incentivize globally beneficial behavior are open questions in this general setting.