According to Bin Hu, exciting research ideas have emerged from combining the control theory and machine learning fields—both of which are important for managing complex systems like self-driving vehicles, delivery drones, humanoid robotics, smart building, and automated healthcare.
For more than 50 years, control theory has guided the creation of sophisticated mathematical tools that underpin modern safety-critical and dynamic systems like commercial aircraft and nuclear power plants. More recently, machine learning algorithms have harnessed the power of massive amounts of data to enable computers to recognize a visual scene, understand written text, or perform an action in the real world.
According to ECE Assistant Professor Bin Hu, exciting research ideas have emerged from combining the control theory and machine learning fields—both of which are important for managing complex systems like self-driving vehicles, delivery drones, humanoid robotics, smart building, and automated healthcare.
“On the one hand, tools from control theory have been tailored for optimization and machine learning applications,” he said. “For example, the feedback control system perspectives of algorithms in optimization and machine learning have led to exciting new results. On the other hand, machine learning techniques have been used to push the boundary of traditional control. For example, reinforcement learning has shown great potential for complex control tasks.”
In the spring of 2021, Hu was awarded a $500,000 NSF CAREER research grant for young faculty to fully explore the interplay between these two fields. He will develop an interdisciplinary approach that merges robust control theory, nonlinear system theory, jump system theory, supervised learning, reinforcement learning, imitation learning, semidefinite programming, and non-convex optimization.
Hu’s interdisciplinary approach is centered around two thrusts. The first thrust focuses on tailoring control theory to unify, streamline, and automate the analysis and design of machine learning algorithms. The second thrust focuses on borrowing recent results in non-convex learning to push control theory beyond the convex optimization regime.
“[My] proposed research covers both ‘control for learning’ and ‘learning for control,’ which deepens the connections of control and learning by showing that the techniques used by each side can be explored to impact the other side,” Hu said.
Another component of NSF CAREER award grants is course and curriculum development. Hu plans to enhance the Interplay Between Control and Machine Learning (ECE 598 ICM) course that he created two years ago. The course covers the cutting-edge research results developed at the intersection of control, optimization, and machine learning, exposing graduate students to an interdisciplinary perspective connecting concepts and techniques across different fields.
The NSF CAREER Award is the agency’s most prestigious award in support of early-career faculty who have the potential to serve as academic role models in both research and education and can advance the mission of their respective department or organization.