ABSTRACT: Designing societal-scale cyber-physical systems such as smart grids and transportation networks is a daunting task. Robustness and safety have always been essential objectives of infrastructure design, but now, in the age of big data and the Internet of Things, data privacy is a necessity. Feedback control is the natural candidate for providing robustness and guaranteeing performance, and aggregation and mechanism design can provide privacy. In this talk I show how these concepts can be combined to provide a unified framework for privacy preserving control.

In the first part of the talk I present a new paradigm for control synthesis called the System Level Approach (SLA). The SLA characterizes the largest known class of tractable distributed control problems and yields convex programs for synthesizing the controllers; moreover, these convex programs scale independently of the size of the network. I show that the notion of spatial and temporal locality is key to achieving such scalability.

In the second half of the talk I describe a framework that incentivizes users to contribute their data, allowing us to build accurate models for control design whilst maintaining user privacy. In keeping with the first half of the talk, locality here plays an important role in determining the maximum achievable privacy levels. Using an energy system case study, I show how performance and privacy trade off against each other and conclude by describing how the frameworks can be extended to a variety of other application areas.

BIO-SKETCH: James Anderson is a Senior Research Scientist in the Department of Computing and Mathematical Sciences at the California Institute of Technology. He obtained a DPhil (PhD) in Engineering Science from the University of Oxford, UK, and the BSc and MSc degrees in Systems Engineering from the University of Reading, UK. Prior to Caltech he held a Junior Research Fellowship in Engineering Science at St John’s College and the University of Oxford, where he was based in the Control Group. His research interests include distributed control, convex optimization, and privacy-preserving computation with applications in autonomy and cyber-physical systems. Since 2012 he has been the lead author of the MATLAB polynomial programming toolbox SOSTOOLS.