My areas of interest are privacy-preserving technologies in cyber-physical systems (CPS). My research to date has focused on developing quantitative tools that combine control theory, machine learning, and optimization to provide fundamental insights to the tradeoff between data utility and privacy, and to further design CPS with improved privacy protection.
The current privacy preservation mechanism in CPS can be categorized into two types - passive and active. Passive mechanisms model the revelation process of private variables to public observables, reactively quantify the privacy leakage, and provide users with “notice and choice”; whereas active mechanisms, also conceptualized as “Privacy by Design” (PBD), proactively incorporate privacy into the design of CPS, and explicitly consider the privacy and performance tradeoff. PBD has been recognized as an essential component and recommended practice of fundamental privacy protection, complied by many countries including US. However, the lack of methodological aspect of PDB also poses considerable difficulty in implementing PBD in practical engineering system design. My research goal is to develop paradigms to incorporate privacy into the design and commissioning phases of CPS. My research realizes this goal from three angles:
Framework to quantify privacy-utility tradeoff
Since privacy is an abstract and subjective concept, a quantitative model of privacy leakage is essential to modeling and analysis of a privacy-preserving system. This project aims to quantify privacy leakage and utility of streaming data collected in CPS, which is, in turn, used to guide the design of optimal privacy-preserving mechanism.
- One-shot design: We develop a tractable framework to study the optimal tradeoff between privacy pursuit and CPS performance.
- Adaptive design: We also consider the privacy mechanism design problem when components of a CPS, such as the environment, sensor, control, communication, etc., are changing and uncertain.
- Free-lunch privacy: We explore and analyze the room for “free-lunch” privacy in a CPS, i.e., the privacy protection without damaging the expected performance.
Privacy-preserving data publication
Spurred on by the mutual benefits of individuals, system operators, and research communities, there is a rising demand for the exchange and publication of the data among various parties. This project aims to design a data publication system specialized for publishing high dimensional data collected in CPS.
Additional areas of interest
- Building performance modeling and simulation
- Occupancy sensing and prediction
- Differential Privacy
- Machine learning in adversarial settings