CAREER: Control theoretic approaches for dynamic and privacy preserving distributed optimization algorithms

University Of California-Irvine, Irvine CA

Investigators

Abstract

Networked systems have undergone advances toward providing efficient solutions to many challenging problems in the modern world. Smart grid operations, smart transportation, and smart healthcare are just a few such networked operations that are envisioned to help us manage our resources efficiently using smarter interactions among their subsystems. However, optimally benefiting from these networked systems will require efficient operational algorithms, many of which involve in-network static or dynamic optimal decision-making. Effective solutions for such algorithms need to be distributed and implementable within the limitations inherent in communication devices. Furthermore, in many applications, these solutions also must have transparent privacy preservation properties to make them acceptable for everyday use. This proposal's research objective is to investigate how automatic control-theoretic tools can be used to develop and analyze distributed dynamic optimization algorithms that respect the restrictions inherent to communication channels and address concerns regarding privacy preservation of participating agents in smart network operation. The results of this research will facilitate the realization and adoption of optimal in-network solutions in many important cyber-physical applications, such as smart grid, sensor networks, and distributed data regression in the healthcare and banking sectors. This project research will be complemented by a multi-tiered educational, mentoring, and outreach plan to train and motivate the next generation of experts who will solve problems that will emerge along with the new paradigms of smart networked systems. The research will be integrated into the University of California, Irvine curriculum through three main avenues: (1) developing new graduate and undergraduate courses, (2) involving students in the PI's Cooperative Systems Lab, and (3) mentoring efforts at the college and high school levels. This project will significantly expand the current state of knowledge on distributed solutions for in-network dynamic optimization algorithm design in three separate directions: design techniques, efficient communication, and transparent privacy preservation characteristics. In terms of design techniques, the proposal will showcase a systematic distributed algorithm design using two time-scale dynamical system concepts from singular perturbation theory. For a robust, active, and smarter communication strategy, event-triggered communication approaches will be used to enable each agent to locally decide when it is necessary to communicate in order to preserve the algorithm's integrity. Finally, to achieve transparent privacy preservation, this project will include development of observability tests to identify a knowledge set that enables eavesdroppers to construct the private data of other agents in the network using sophisticated observers, such as neuro-observers. The project will also include design of solutions for resistance of these intrusions by developing tools and methods to create augmentations that induce privacy preservation in distributed dynamic optimization algorithms.