Research Overview

I am broadly interested in the areas of mobile Cyber-Physical Systems (CPS) and Data Science by a technical integration of data, humans, and systems to achieve an ultimate goal of Human-System Synergy.

My current work is focsed on two directions:
  • human behavior learning driven by rich data collected from urban physical systems;
  • applications based on human behavior knowledge.

Selected Topics

Socially Informed Services Conflict Governance

project image

Smart city services are deeply embedded in modern cities aiming to enhance various aspects of citizens’ lives. However, underlying expected or unexpected couplings among services due to complex interactions of social and physical activities are under-explored, which lead to potential conflicts. Funded by NSF using City of Newark in New Jersey as a testbed, this project aims to develop ways of reducing conflicts for ensuring social inclusion and equity of city services to achieve a "harmony" among various city services.[Website]

A Citywide On-demand Delivery System with 100 thousand couriers and 7.3 million customers

We studied the whole process of an on-demand delivery system Eleme in a city. Especially, we (i) analyzed the couriers' delivery behavior; (ii) predicted the couriers' indoor movement; (iii) deployed and operated 12 thousand BLE beacon devices. The results were published in MobiCom'20 and NSDI'21.

A Citywide Heterogeneous Systems with 50 thousand vehicles, 8-line subways, 3 million users

project image

In this project, we studied the interaction between multiple urban systems, including a heterogeneous fleet system, a subway system, and a cellular network system. Especially, we are interested in vehicular mobility, traffic patterns, and data usage behavior. The results were published in UbiComp'20, UbiComp'19 and UbiComp'18.

A Statewide Highway System with daily 2 million vehicles

In this project, we studied the statewide vehicular mobility on highways based on an electronic toll collection system. We are especially interested in predicting the travel time and real-time locations of vehicles without using GPS information. The results were published in MobiCom'19 and UbiComp'18.

A Nationwide Vehicular System with 1.5 million vehicles in 50 cities

In this project, we studied phenomena sensed by vehicles across multiple cities. Especially, we are interested in applying transfer learning to solve the data scarcity problem such as driving risk inference and crowd mobility modeling.