OPTIMAL CONTROL AND LEARNING FOR DYNAMICAL SYSTEMS UNDER ADVERSARIAL ATTACKS
The robust learning of dynamical systems is crucial for safety-critical applications, such as power systems and autonomous systems. The control theory has a rich literature on system identification and optimal control in the case when the system is subject to non-adversarial and mostly Gaussian disturbance. However, there is a pressing need to develop learning and control techniques for systems whose inputs are under adversarial attacks, meaning that the system operates in a hostile environment. The main challenge is that the classic results relying on closed-form solutions for least-square estimators and LQR/LQG are no longer valid, and it is essential to design non-smooth estimators and controllers with no closed-form solutions in presence of adversarial attacks. In this talk, we discuss the recent advances in the area and focus on the problem of learning an unknown nonlinear dynamical system subject to adversarial disturbance/input. We develop a non-smooth estimator and show that the correct dynamics of the system can be learned in finite time no matter how severe the attack is as long as the learning period is longer than some threshold. We then study optimal control for systems in hostile environments.
Date and Time
Location
Hosts
Registration
-
Add Event to Calendar
Loading virtual attendance info...
- Starts 24 September 2025 04:00 AM UTC
- Ends 18 November 2025 09:00 PM UTC
- No Admission Charge
Speakers
Dr.Lavaei
OPTIMAL CONTROL AND LEARNING FOR DYNAMICAL SYSTEMS UNDER ADVERSARIAL ATTACKS
The robust learning of dynamical systems is crucial for safety-critical applications, such as power systems and autonomous systems. The control theory has a rich literature on system identification and optimal control in the case when the system is subject to non-adversarial and mostly Gaussian disturbance. However, there is a pressing need to develop learning and control techniques for systems whose inputs are under adversarial attacks, meaning that the system operates in a hostile environment. The main challenge is that the classic results relying on closed-form solutions for least-square estimators and LQR/LQG are no longer valid, and it is essential to design non-smooth estimators and controllers with no closed-form solutions in presence of adversarial attacks. In this talk, we discuss the recent advances in the area and focus on the problem of learning an unknown nonlinear dynamical system subject to adversarial disturbance/input. We develop a non-smooth estimator and show that the correct dynamics of the system can be learned in finite time no matter how severe the attack is as long as the learning period is longer than some threshold. We then study optimal control for systems in hostile environments.
Biography:
Javad Lavaei is an Associate Professor at the University of California, Berkeley. He was an Assistant Professor at Columbia University from 2012 to 2015. He received the Ph.D. degree in Control and Dynamical Systems from the California Institute of Technology in 2011 and received the Milton and Francis Clauser Doctoral Prize for the best university-wide Ph.D. thesis. Javad Lavaei has won several awards, including the Presidential Early Career Award for Scientists and Engineers given by the White House, DARPA Young Faculty Award, Office of Naval Research Young Investigator Award, Air Force Office of Scientific Research Young Investigator Award, NSF CAREER Award, DARPA Director's Fellowship, Office of Naval Research's Director of Research Early Career Grant, and Google Faculty Award. Javad Lavaei is a senior editor of the IEEE Systems Journal and has served on the editorial boards of the IEEE Transactions on Automatic Control, IEEE Transactions on Control of Network Systems, IEEE Transactions on Smart Grid, and IEEE Control Systems Letters. He has also served on the conference editorial boards of the Control Systems Society and European Control Association. Javad Lavaei is a recipient of 2015 INFORMS Optimization Society Prize for Young Researchers, 2016 Donald P. Eckman Award given by the American Automatic Control Council, 2016 INFORMS ENRE Energy Best Publication Award, 2017 SIAM Control and Systems Theory Prize, 2020 Journal of Global Optimization Best Paper Award, and 2022 Antonio Ruberti Young Researcher Prize. He has received over 10 best paper awards from the Control Systems Society, INFORMS, and Power & Energy Society. Javad Lavaei is a Fellow of IEEE.
Email:
Address:INDUSTRIAL ENGINEERING AND OPERATIONS RESEARCH, COLLEGE OF ENGINEERING, UC BERKELEY, Berkeley, California, United States, 94720-1777