Dissertation Talk: Accountable Data Fusion and Privacy Preservation Techniques in Cyber-Physical Systems
Seminar: Departmental: Dissertation Talk: EE | September 25 | 490H Cory Hall
Ruoxi Jia, UC Berkeley
With the deployment of large sensor-actuator networks, Cyber-Physical Systems (CPSs), such as smart buildings, smart grids, and transportation systems, are producing massive amounts of data often in different forms and quality. These data are in turn being used collectively to inform decision-making of the entities that engage with the CPSs. The impact of these systems on people's lives has led to a strong call for accountability of system decisions made based upon various data sources. The collection, analysis, and dissemination of these data also present a privacy risk that needs to be addressed.
In the first part of this talk, I will discuss a principled way to characterize the value of different data sources for any given data-enabled decisions or services, and provide efficient algorithms for data valuation. This not only enables us to better understand black-box predictions through the lens of training data but allows for the fair allocation of the profit generated from a model that is built with data from cooperative entities. We use the proposed data value notion to develop an effective data sanitization mechanism, which can effectively screen off low-quality or even adversarial data instances from the training set.
In the second part of the talk, I will focus on the problem of incorporating privacy as an active engineering constraint into the CPS design and operation. I will discuss a privacy metric inspired by information theory and provide algorithms to optimize the privacy mechanism for a given system or co-design the privacy mechanism and system control. While the algorithms and techniques introduced can be applied to many CPSs, we will mainly focus on the implications for smart buildings.