research
My research focuses on developing optimization, control, and learning algorithms for multi-agent systems. I am particularly interested in applications to aerial mobility (e.g., air taxis, delivery drones) and surface transportation systems(e.g., electric and autonomous vehicles, shared mobility platforms). My current research projects are listed below.
Privacy-preserving optimization and control:
Competition or privacy considerations can limit the data available for optimal control and decision-making. However, by using tools from cryptography, online learning, differential privacy, and distributed algorithms, I am interested in (a) characterizing the fundamental performance limits of our algorithms under privacy constraints, (b) leveraging domain expertise and the problem structure to design algorithms that can achieve the optimal performance, and (c) inducing policy changes by demonstrating, using real-world data, that privacy and efficiency may not always be at odds if the system is designed appropriately.
Related publications: 1, 2, 3, 4
Learning-enabled scheduling and planning:
Many scheduling and planning problems are posed as network flow problems, large linear programs, mixed-integer linear programs, or even integer programs. Solving these optimization problems for large-scale systems in real-time has been a challenge for a long time, and led to the development of approximation algorithms or special optimization algorithms that leverage properties of the specific problem instance. I am interested in leveraging the recent advances in machine learning to tackle this classic challenge. In particular, I focus on (a) using reinforcement learning to solve computationally intractable optimization problems as a sequence of learning-guided tractable sub-problems, and (b) using data-driven approaches for improved forecasting of uncertain parameters in stochastic optimization problems.
Algorithms for equitable scheduling and planning:
While the need for equitable systems is widely acknowledged, their deployment in practice is rather limited due to poor alignment of incentives, and the associated loss in system performance. Our research focuses on (a) using algebraic and geometric tools to identify the conditions under which equity can be achieved without any trade-offs, and when it may be impossible to do so, i.e., the problem is ''equity-hard'', (b) studying how incentives can be engineered (e.g., using tokens, karma-credits, etc.) to steer rational agents towards equitable behavior and (c) developing algorithms to solve large-scale non-linear optimization problems that result from introducing fairness.
Related publications: 7, 8, 9, 10
Robust and resilient network operations:
We require critical infrastructure systems (e.g., communication, power, roads, air traffic) to operate robustly under off-nominal conditions and recover swiftly from disruptions. However, complex interactions, time-varying network topologies, multi-layered network structures, and noisy data make it challenging to develop control policies for optimal network performance under off-nominal conditions. I am interested in using tools from graph theory, switched-systems theory, machine learning, and optimal control to a) predict and characterize network disruptions, b) develop scheduling and routing policies that are robust to disruptions, and c) design data-driven controllers to recover swiftly after disruptions.