Thank you for your interest in the DCL Student Symposium! The Symposium is an annual event which brings together members of the Georgia Tech community working in the area of systems and controls to network and share their research. The Symposium is entirely organized by DCL students, and it is our hope that connections between students working in labs across the Institute will be forged at this event.

This year, we are honored to have Dr. Jorge Cortés and Dr. Quanyan Zhu as our plenary speakers. They will also join our new DCL faculty members Dr. Matthew Hale and Dr. Maegan Tucker later in the day for a panel discussing career paths in controls. Honoring the spirit of the Symposium, we will also have multiple sessions of student contributions throughout the day featuring both spotlight talks and posters. Of course, coffee and snacks will be provided along with a catered lunch. 

In this rest of this document, you will find a schedule of the day along with details of the speakers and presentations. We look forward to seeing you this Friday, February 23rd at the Symposium!

Organizing Committee, DCL Student Symposium 2024


23rd of February 2024
09:00 – 09:15Opening Remarks
09:15 – 10:00Student Spotlight Talks I
10:00 – 11:00Student Poster Session + Coffee Break
11:00 – 12:00Keynote Speaker I – Jorge Cortés
12:00 – 13:00Lunch
13:00 – 13:45Student Spotlight Talks II
13:45 – 14:00Break
14:00 – 15:00Keynote Speaker II – Quanyan Zhu
15:00 – 15:15Break
15:15 – 15:45Student Spotlight Talks III
15:45 – 16:00Break
16:00 – 16:55Panel Talk
16:55 – 17:00Closing Remarks


The safe gradient flow: a system-theoretic approach to anytime constrained optimization through control barrier functions

Abstract: This talk is motivated by problems where the solution to a constrained optimization problem is used to regulate a physical process modeled as a dynamically evolving plant (e.g., to provide setpoints, specify optimization-based controllers, or steer the system toward an optimal steady state).  This type of problem arises in multiple application areas, including safety-critical control, power networks, traffic networks, and network congestion.  A paradigmatic example would be the use of CBF-based quadratic programs for controller synthesis in robotics.  We are interested in situations where the problem incorporates constraints which, when violated, threaten the safe operation of the physical system.  In such cases, the algorithm that solves the optimization must be anytime, meaning that it is guaranteed to return a feasible point even when terminated before it has converged to a solution.  We introduce a novel control-theoretic algorithm for solving constrained nonlinear programs that combines continuous-time gradient flows to optimize the objective function with techniques from control barrier functions to maintain forward invariance of the feasible set.  We discuss the remarkable properties of the resulting closed-loop system, which we term safe gradient flow, regarding regularity, stability, convergence, and invariance.  Comparisons with other continuous-time methods for optimization illustrate the advantages of the safe gradient flow.

Bio: Jorge Cortes is a Professor and Cymer Corporation Endowed Chair in High Performance Dynamic Systems Modeling and Control in the Department of Mechanical and Aerospace Engineering, University of California, San Diego.  He is the author of “Geometric, Control and Numerical Aspects of Nonholonomic Systems” (New York: Springer-Verlag, 2002) and co-author of “Distributed Control of Robotic Networks” (Princeton: Princeton University Press, 2009).  He is a Fellow of IEEE, SIAM, and IFAC.  He has co-authored papers that have won the 2008 and the 2021 IEEE Control Systems Outstanding Paper Award, the 2009 SIAM Review SIGEST selection from the SIAM Journal on Control and Optimization, the 2012 O. Hugo Schuck Best Paper Award in the Theory category, and the 2019 and 2023 IEEE Transactions on Control of Network Systems Outstanding Paper Award.  At the IEEE Control Systems Society, he has been a Distinguished Lecturer (2010-2014), an elected member (2018-2020) of the Board of Governors, and Director of Operations (2019-2022) of its Executive Committee.  His research interests include distributed control and optimization, network science and complex systems, resource-aware control and coordination, distributed decision making and autonomy, network neuroscience, and multi-agent coordination in robotic, power, and transportation networks.

(Personal Website: http://terrano.ucsd.edu/jorge)


Strategic Learning for Cyber-Physical Resilience: The Confluence of Games, Control, and Learning

Abstract: The rapid growth in the number of devices and their connectivity has enlarged the attack surface and made cyber systems more vulnerable. As attackers become increasingly sophisticated and resourceful, mere reliance on traditional cyber protection. Resilience provides a new security paradigm that complements inadequate protection with resilience mechanisms. A resilient mechanism adapts to the threats and uncertainties in real-time and strategically responds to them to maintain the critical functions of the systems. In this talk, we discuss several learning paradigms that enable resilience for both IT and OT systems, emphasizing their roles in defending against three primary types of vulnerabilities: posture-related, information-related, and human-related vulnerabilities. Within this framework, we explore three application domains—moving target defense, defensive cyber deception, and assistive human security technologies. The learning algorithms also have vulnerabilities themselves. We discuss the future challenges of strategic learning for cyber security and resilience.

Bio: Quanyan Zhu received B. Eng. in Honors Electrical Engineering from McGill University in 2006, M. A. Sc. from the University of Toronto in 2008, and Ph.D. from the University of Illinois at Urbana-Champaign (UIUC) in 2013. After stints at Princeton University, he is currently an associate professor at the Department of Electrical and Computer Engineering, New York University (NYU). He is an affiliated faculty member of the Center for Urban Science and Progress (CUSP) and Center for Cyber Security (CCS) at NYU. He is a recipient of several awards, including NSF CAREER Award and INFORMS Koopman Prize. He is an Associate Editor of IEEE Transactions on Aerospace and Electronic Systems. He currently serves as the technical committee chair on security and privacy for the IEEE Control Systems Society. His current research interests include game theory, machine learning, cyber deception, network optimization and control, cyber and physical system resilience. His research has been funded by DARPA, IARPA, ARO, NSF, DHS, and DOE. He is a co-author of several recent books: Cognitive Security: A System Scientific Approach (with L. Huang), Cyber-Security in Critical Infrastructures: A Game-Theoretic Approach (with S. Rass, S. Schauer, and S. König), Game Theory for Cyber Deception (with J. Pawlick), and Cybersecurity in Robotics (with S. Rass, B. Dieber, V. M. Vilches). 

(Personal Website: https://wp.nyu.edu/quanyan/)


Session 1 (09:15 – 10:00)

Jason Ye

(Advisor: Dr. Joseph K. Scott)

Improving the Efficiency of Branch-and-Bound for Global Dynamic Optimization

This talk deals with my work in improving the computational efficiency of using the technique of branch-and-bound for solving global dynamic optimization (GDO) problems. Such problems are important in modeling and optimizing multiple processes, ranging from batch reactions to motion planning, where dynamics are involved. Despite branch-and-bound’s ability to solve GDO problems with up to around 5 state and 10 decision variables as input in around 3 hours, real GDO problems such as the optimization of batch reactions at the industrial scale often involve hundreds of such variables. I will present improvements to a key step in the branch-and-bound algorithm and show that these improvements can enhance the speed of branch-and-bound anywhere from 1.5 times to over 30 times, thus bringing us much closer to the goal of solving real GDO problems.

Mohannad Alkhraijah

(Advisor: Dr. Daniel K. Molzahn)

Detecting Shared Data Manipulation in Distributed Optimization Algorithms

Distributed optimization algorithms allow multiple agents to collaboratively solve large-scale optimization problems. Using distributed optimization to solve real-world problems has many advantages in terms of scalability, reliability, and privacy. However, the reliance on communication makes distributed optimization algorithms vulnerable to cyberattacks and data manipulation. In this talk, we present methods to detect data manipulation on distributed algorithms based on theoretical results and adversarial neural network framework.

Biswadeep Chakraborty 

(Advisor: Dr. Saibal Mukhopadhyay)

Leveraging Evolution Strategies in Heterogeneous Recurrent Spiking Neural Networks for Dynamic Control

Leveraging the intrinsic dynamism of recurrent spiking neural networks (RSNNs), which mirror the biological nervous system’s complexity, holds transformative potential for artificial general intelligence. Yet, surrogate gradient-based training, the common approach for RSNNs, falls short in precision and compatibility with neuromorphic hardware. To navigate these challenges, we introduce an Spike Timing Dependent Plasticity (STDP)-based evolving connectivity (STDP-EC) framework, which pivots from conventional weight tuning to an inference-centric methodology. The STDP-EC framework pioneers the path for energy-conservative neuromorphic implementations and paves the way for the next generation of neuromorphic computing devices.

Session 2 (13:00 – 13:45)

Trent Schreiber 

(Advisor: Dr. Yang Wang)

Finite Element Model Updating using Primal-Relaxed Dual Global Optimization Algorithm

Finite element (FE) modeling has become a powerful tool in predicting the response of various engineering structures. However, predictions from the numerical model often differ from in-situ experimental measurements due to numerous approximations and inaccuracies in the model. The in-situ experimental data obtained from the as-built structure can be used to update selected model parameters to obtain a more accurate FE model that truly reflects the behavior of the as-built structure. This research investigates FE model updating by the modal property difference approach using eigenvalues and eigenvectors. The modal property difference approach is a nonconvex optimization problem, for which generic solvers cannot guarantee global optimality. However, the problem can be reformulated into a biconvex problem so that the global optimum can be found using a primal-relaxed dual (P-RD) decomposition approach.

Nejat Tukenmez 

(Advisor: Dr. Kyriakos G. Vamvoudakis)

Intermittent Learning Framework on Micro drones”

An online learning scheme, which promotes the intermittent updated control policy by lagging the reinforcement signals, has been introduced. In addition, we support the proposed framework stability and optimality analysis. Finally, the framework has been validated both in simulation environment and on a real quadrotor system in which all learning and control computations are conducted in real-time to show the efficacy of the learning mechanism.

William Warke 

(Advisor: Dr. Matthew Hale)

Computation-Aware Bearings-Only Target Localization and Circumnavigation in 2D

We present a novel framework for computation-aware guidance, navigation, and control (GNC) that allows a mobile autonomous agent to act on intermediate iterates of its computations as it moves. In particular, we consider an iterative optimization algorithm running onboard the agent, and each iterate of the algorithm is the most recent estimate of an optimum, while a convergence rate bounds the distance from that iterate to an optimum. The agent implements a GNC system that acts on its most recent iterate in a way that is modulated by its convergence rate. Then, we apply this framework to solve the problem of bearings-only target localization and circumnavigation in 2 dimensions in a computation-aware manner, and we bound the error in control performance that is introduced by using online computations.

Session 3 (15:15 – 15:45)

Nguyen Hoang

(Advisor: Dr. Siva Theja Maguluri)

Stochastic Approximation for Nonlinear Discrete Stochastic Control: Finite-Sample Bounds

We consider a nonlinear discrete stochastic control system, and our goal is to design a feedback control policy in order to lead the system to a prespecified state via a stochastic approximation viewpoint. In this paper, we provide finite-sample convergence bounds whenever a Lyapunov function is known for the continuous system in four different cases based on whether the Lyapunov function for the continuous system gives exponential or sub-exponential rates and based on whether it is smooth or not. Our proof relies on constructing a Lyapunov function for the discrete system based on the given Lyapunov function for the continuous system and we present numerical experiments corresponding to the various cases, which validate the rates we establish.

Shaan Haque

(Advisor: Dr. Siva Theja Maguluri)

Tight Finite Time Bounds of Two-Time-Scale Linear Stochastic Approximation with Markovian Noise

Stochastic approximation is an iterative algorithm to find the fixed point of the operator given its noisy samples. It has been widely applied in machine learning and reinforcement learning. In many settings, it is implemented in a two-time scale manner where one time scale estimates an auxiliary parameter of the system while the other uses this estimate to find the fixed point. Recently, there has been an increasing focus on how fast these algorithms converge to the fixed point. In this paper, we analyze two-time scale linear stochastic approximation when the noise follows Markovian behavior and characterize the rate of the convergence of the algorithm.


Bo Chen

Differential Privacy in Cooperative Multiagent Planning

Evanns G. Morales-Cuadrado

Newton-Raphson Flow for Aggressive Quadrotor Tracking Control

Nejat Tukenmez

Intermittent Reinforcement Learning Framework for Agile Unmanned Aerial Vehicles

Young In Kim

A Streamlined Heuristic for The Problem of Min-Time Coverage in Constricted Environments

Biswadeep Chakraborty

Evolution Strategies in Heterogenous Recurrent Spiking Neural Network for Dynamical Control

Zishun Liu

Data-Driven Online Optimal Control with Nonstochastic Disturbance

Luke Baird

Runtime Assurance from Signal Temporal Logic Safety Specifications on a Miniature Autonomous Blimp

Calvin Hawkins

Multistream Anomaly Detection

Trent Schreiber

Finite Element Model Updating using Primal-Relaxed Dual Global Optimization Algorithm

Tyler Rome

Manual and Automated Obstacle Field Navigation of System with Changing Oscillatory Modes

Steven Crouse

Dual Kalman Filter for Detecting Mixing Faults in Nuclear Waste Processing

Alison Jenkins

Input Shaping for Motor Control for Haptics and Human Perception