Organizers: Aranya Chakrabortty (National Science Foundation)
Time: May 28, Friday, 10:15-12:15 AM CDT
The purpose of the tutorial session is to review recent accomplishments and emerging opportunities in control systems from the perspective of U.S. National Science Foundation (NSF) CAREER award recipients. The CAREER program is the most prestigious award in support of early-career faculty who have potential to serve as academic role models in research and education and to lead advances in the mission of their department or organization. The CAREER awardees will discuss the latest breakthrough research in their respective field and key research opportunities in systems and control.
Title: Differential Privacy in Feedback Systems
Presenter: Matthew Hale (Univ. of Florida)
Time: 10:15-10:35 AM CDT
Recent developments in cyber-physical systems and the Internet of Things take a data-driven approach to control and feedback. When data belongs to users, it can reveal sensitive details of their personal lives, and thus the engineering need for data is at odds with individuals' desire to conceal it. Recently, differential privacy has emerged as a formalism that can simultaneously safeguard data and preserve its usefulness, though substantial efforts are still required to bring its protections to control systems while ensuring that systems continue to perform well. This talk will review recent breakthroughs on differential privacy in LQ control and policy synthesis for MDPs, along with future research directions and outstanding questions in private control.
Title: Adaptive Reinforcement-Learnings
Presenter: Siddhartha Banerjee (Cornell)
Time: 10:35-10:55 AM CDT
Reinforcement learning (RL) is a natural paradigm for sequential decision-making which has enjoyed much recent success; however most state-of-the-art RL algorithms have high data/memory/energy requirements, which make them impractical for many applications (in particular, in real-time control). I will describe a recent line of work, where we develop new RL algorithms based on a novel data-driven adaptive discretization approach. Our algorithms are based on a unified framework for discretizing the state-action space in a way which zooms into promising regions (either in terms of large immediate rewards, or leading to high future rewards). Our techniques apply to both model-free and model-based RL paradigms, and in both cases, lead to policies with performance guarantees that adapt to the underlying problem complexity. Empirically, our algorithms sample more efficiently and require much less memory and computation at every iteration, and yet, perform as well as the state-of-the-art. (Joint work with Sean Sinclair and Christina Yu).
Title: Roles of Directed Information in Networked Control Systems Theory and Beyond
Presenter: Takashi Tanaka (UT Austin)
Time: 10:55-11:15 AM CDT
Directed information, an information-theoretic concept broadly used for causality analysis, provides a rigorous measure of “information flow” in networked systems. In networked control systems theory, directed information can be used to quantify the feedback data rate and to derive a fundamental limitation of control-communication co-design. In this presentation, we summarize prominent roles of directed information in systems and control, ongoing research efforts toward the unification of control and information theory, and existing challenges. We also discuss emerging applications of directed information beyond the conventional scope of networked control systems theory, such as strategic perception and task-dependent information processing for autonomous agents in dynamical environments.
Title: Privacy preserving control and optimization for cyber-physical systems
Presenter: Minghui Zhu (Pennsylvania State University)
Time: 11:15-11:35 AM CDT
Cyber-physical systems (CPSs) consist of a large number of geographically dispersed entities and distributed data sharing is necessary to achieve network-wide goals. However, distributed data sharing raises the significant concern that private information of legitimate entities could be leaked to unauthorized entities. The privacy issues of CPSs have been exposed in various areas; e.g., occupancy-based heating, ventilation, and air conditioning (HVAC) control systems, urban sensing networks, smart meters and drones. Most existing privacy preserving techniques solely focus on the cyber space but ignore the physical world. Hence, they alone may not be adequate to ensure CPS privacy.
In this talk, we will discuss how to leverage control theory to complement existing privacy preserving techniques. More specifically, we will consider the problem that a data requester demands a group of agents to release real-time outputs of a linear dynamic network, and, on the other hand, the agents aim to prevent the data requester from inferring their private information using the released data. We will present a scheme of adding feedback perturbations into system inputs and outputs such that (i) network privacy is protected; (ii) system utilities; e.g., controllability, are maintained; and (iii) costs induced by the perturbations are minimized. A major merit of our perturbation scheme is that the added perturbations are diminishing as the system approaches the origin. The efficacy of the proposed technique is verified by a case study on an HVAC system.
Title: NSF Programs in Control, Robotics, Smart-Grid, and Cyber-Physical Systems
Presenter: Aranya Chakrabortty (National Science Foundation)
Time: 11:35-11:55 AM CDT
The goal of this talk is to provide an update on National Science Foundation (NSF) funding opportunities in the area of Dynamic Systems, Control and Networked Systems research and education. Research projects in power systems with renewable energy integration, power electronics, and open-access testbeds will be presented. The presentation will include NSF programs in Cyber-Physical Systems (CPS), and National Robotics Initiative. The CPS program brings together researchers from computations, communications, and control disciplines to address important engineering problems. The presentation will include recent activities at NSF in Smart and Connected Communities.