Research Interests

My research is on how to use data to make better decisions in the operation of large-scale cyber-physical systems such as power and transportation networks. I have worked on optimization algorithms that turn data into useful actions (i.e., data-driven optimization). Because a lot of data used for decision-making are often coming from human users, I have also worked on optimization algorithms that preserve the privacy of individual users. More recently, motivated by the advances in self-driving cars, I have been looking into how autonomy may change the future of urban transportation.

I. Data-Driven Decision Making

How do we turn data into useful actions?

Decision Making with Statistics from Sampled Data

How do we use data for decision making when only statistics from sampled data are available?

Distributionally robust optimization is a framework that unifies robust optimization and stochastic programming. Compared to stochastic programming, the framework is based on stochastic modeling, but it only uses statistics from sampled data that are easier to obtain than a full stochastic model. The key assumption is that the underlying probability distribution belongs to a set of distributions (defined by the statistics), similar to conventional robust optimization. Naturally, distributionally robust optimization requires solving an infinite-dimentional optimization problem and is not amenable to numerical solutions. I have developed tractable numerical solutions for a very general class of distributionally robust optimization problems. The method has been applied to the problem of energy storage assignment in a power network in the presence of uncertain wind power generation.

  1. Shuo Han, Molei Tao, Ufuk Topcu, Houman Owhadi, and Richard M. Murray, “Convex optimal uncertainty quantification,” SIAM Journal on Optimization, vol. 25, no. 3, pp. 1368–1387, 2015. [pdf]
  2. Shuo Han, Ufuk Topcu, Molei Tao, Houman Owhadi, and Richard M. Murray, “Convex optimal uncertainty quantification: Algorithms and a case study in energy storage placement for power grids,” in American Control Conference, 2013. Best Student Paper Finalist [pdf]

Data-Driven Resource Allocation for Dynamical Networks

How do we allocate resource in a dynamical network by knowing only time-series data from sensors?

Spreading processes on a graph can be used to model phenomena such as cascading failures in a power network and virus spreading within a computer network. For many applications, we would like to control the rate of spreading within a given budget. I have worked on the problem of controlling spreading processes in an unknown network based on time series data describing the evolution of the spreading process collected from observations. A common approach is to identify from observed data a single network that is subsequently used for resource allocation. Such kind of identification problem is often ill-posed, so that additional prior knowledge is needed. Instead of determining the appropriate prior knowledge, my formulation tries to compute a robust allocation for the worst-case network that is consistent with observed data. I have developed an efficient numerical solution that can be used to allocate resources for a network consisting of hundreds of nodes. I have also conducted an empirical study of the solution method on epidemic control within a geographic region.

  1. Shuo Han, Victor M. Preciado, Cameron Nowzari, and George J. Pappas, “Data-driven network resource allocation for controlling spreading processes,” IEEE Transactions on Network Science and Engineering, vol. 2, no. 4, pp. 127–138, 2015. [pdf]

II. Privacy in Cyber-Physical Systems

How should we design cyber-physical systems that respect the privacy of users?

Private Resource Allocation

How do we handle private data from users for decision making while preserving user privacy?

Many decision making problems in the operation of urban infrastructures, such as electric vehicle charging, demand response, and traffic congestion reduction, can be viewed as a resource allocation problem with user participation. In these problems, a central entity needs to compute an allocation that maximizes social welfare based on user preferences/requirements on the resources. These preferences or requirements often contain private and sensitive data that individual users wish to keep undisclosed from the public.

I have worked on privacy-aware resource allocation, which involves techniques in differential privacy and distributed optimization. The result allows computing a nearly socially optimal resource allocation without compromising the privacy of participating users. I have demonstrated the effectiveness of this privacy-aware resource allocation framework in the context of electric vehicle charging. In addition, the privacy-preserving charging algorithm also guarantees (approximate) truthfulness of the users. This means that users (who are assumed selfish) have little motivation of using a false specification in the interest of their own.

  1. Shuo Han, Ufuk Topcu, and George J. Pappas, “Differentially private distributed constrained optimization,” IEEE Transactions on Automatic Control, vol. 62, no. 1, pp. 50–64, 2017. [pdf]
  2. Shuo Han, Ufuk Topcu, and George J. Pappas, “An approximately truthful mechanism for electric vehicle charging via joint differential privacy,” in American Control Conference, 2015. [pdf]
  3. Shuo Han, Ufuk Topcu, and George J. Pappas, “Differentially private distributed protocol for electric vehicle charging,” in Annual Allerton Conference on Communication, Control, and Computing, 2014. [pdf]
  4. Fragkiskos Koufogiannis, Shuo Han, and George J. Pappas, “Computation of privacy-preserving prices in smart grids,” in IEEE Conference on Decision and Control, 2014. [pdf]
  5. Shuo Han, Ufuk Topcu, and George J. Pappas, “Differentially private convex optimization with piecewise affine objectives,” in IEEE Conference on Decision and Control, 2014. [pdf]

Local Privacy in Distributed Systems

How do we ensure local privacy of users in the absence of a trusted central aggregator?

While differential privacy is able to provide strong privacy guarantees for resource allocation where the user preferences/requirements contain sensitive data, the resulting solution can be too conservative to ensure good system performance in the absence of a trusted central mediator or when the communication links between the trusted mediator and users can be potentially compromised. In these situations, it may be necessary to ensure local privacy at the user level so that the data provided by users are no longer sensitive (but still useful) even before aggregation. One direction that I have been pursuing is the notion of information-theoretic privacy, which quantifies the level of privacy using mutual information between the input and output of the private algorithm. In particular, I consider the setting of processing data streams in which the private algorithm can be viewed as a dynamical system whose input is the data streams under processing. One potential application is private smart metering with energy storage, where the goal is to mask the energy consumption patterns of appliances by using energy storage devices installed at home as a buffer.

  1. Shuo Han, Ufuk Topcu, and George J. Pappas, “Event-based information-theoretic privacy: A case study of smart meters,” in American Control Conference, 2016.

III. Autonomy-Enabled Urban Transportation

How do we use autonomous vehicles to bring more efficient and reliable transportation for cities?

Data-Driven Rebalancing of Vehicles for Ridesharing

How do we make use of data to rebalance vehicles to meet passenger demand?

In ridesharing, an important metric of service quality is the average wait time of passengers. The key to reducing the average wait time is to dynamically rebalance the vehicles to match passenger demand at different locations. The goal of this research is to study how to make use of predictions of future passenger demand from historical data in order to improve the service quality of automated ridesharing. We have developed a data-driven online scheduling algorithm using the theory of distributionally robust optimization, which allows us to address the uncertainties in demand prediction. We also showed that the robust optimization problem can be converted to an equivalent form for which numerically efficient solutions are available. We evaluate the performance of the data-driven vehicle balancing framework based on four years of taxi trip data from New York City.

  1. Fei Miao, Shuo Han, Shan Lin, Qian Wang, John Stankovic, Abdeltawab Hendawi, Desheng Zhang, Tian He, and George J. Pappas, “Data-Driven Robust Taxi Dispatch under Demand Uncertainties,” IEEE Transactions on Control Systems Technology, vol. 27, no. 1, pp. 175–191, 2019. [pdf]
  2. Fei Miao, Shuo Han, Abdeltawab Hendawi, Mohamed E. Khalefa, John A. Stankovic, and George J. Pappas, “Data-driven distributionally robust vehicle balancing with dynamic region partition,” in ACM/IEEE International Conference on Cyber-Physical Systems, 2017.
  3. Fei Miao, Shuo Han, Shan Lin, John A. Stankovic, Desheng Zhang, Sirajum Munir, Hua Huang, Tian He, and George J. Pappas, “Taxi Dispatch with Real-Time Sensing Data in Metropolitan Areas: A Receding Horizon Control Approach,” IEEE Transactions on Automation Science and Engineering, vol. 13, no. 2, pp. 463–478, 2016. [pdf]
  4. Fei Miao, Shuo Han, Shan Lin, and George J. Pappas, “Taxi dispatch under model uncertainties,” in IEEE Conference on Decision and Control, 2015. [pdf]

Comparison between Direct vs. Price-Based Vehicle Rebalancing

How much efficiency can we gain by switching to autonomous vehicles in ridesharing?

Ridesharing services often require constant rebalancing of vehicle supplies in order to meet passenger demand across the transportation network. Current ridesharing services are mostly provided by human drivers, where rebalancing is controlled by setting price differences between service regions so that drivers are incentivized to relocate to regions with higher prices. In recent years, there has been a growing interest in introducing autonomous cars into ridesharing. Since autonomous cars can receive commands from a central authority and be directly dispatched, they provide more flexibility in terms of control and can potentially improve the efficiency of rebalancing. In our work, we proposed a metric that quantifies the efficiency gain of automated ridesharing by comparing the total amount of vehicle flow required for rebalancing. The proposed metric is independent of the actual demand and only relies on properties of the transportation network. We have also developed a set of numerical tools for computing the metric, which are applied to a case study of the Washington, DC metropolitan area.

  1. Shuo Han, Ufuk Topcu, and George J. Pappas, “Quantification on the efficiency gain of automated ridesharing services,” in American Control Conference, 2017.