CSL researchers launch a new institute to use data science to improve decision-making of dynamical systems

A new institute at the University of Illinois Urbana-Champaign plans to break down the boundaries between the fields of data science and dynamical systems. Housed at the interdisciplinary Coordinated Science Laboratory, the Illinois Institute for Data Science and Dynamical Systems (iDS²) will bring together researchers from across campus to develop new foundational principles, which will allow for better decision-making within dynamical systems, such as smart cities or autonomous vehicles.

“Our goal is to establish theoretical foundations at the intersection of data science and dynamical systems, which are only now beginning to be systematically studied and explored,” said CSL’s Max Raginsky, the institute’s principal investigator and an associate

professor of electrical and computer engineering. “This will require an incorporation of techniques, theories, and principles from machine learning, control, game theory, optimization, and statistics.”

Data science is already a part of many people’s daily lives. Whenever a person uses an app on their cell phone, they are providing data, or feedback, to the app creator. This process of data exchange is a dynamical system. As society continues to move into the age of dynamical systems with the addition of autonomous vehicles, smart cities, personalized medicine, and artificial intelligence, dynamical systems will become more ingrained into everyday life.

A dynamical system is any system that changes over time; this might include a chess-playing computer that changes its strategy based on an opponent’s move, delivery drones that take feedback from the current weather to change how they fly, or an autonomous vehicle that senses a pothole in the road and maneuvers around it. To improve decision making of these systems, the researchers will use machine learning -- which facilitates computers interacting with the world, learning from the interaction, and using that information to influence future decisions – to create better tools.

Raginsky is leading the project along with CSL faculty Niao He (industrial and enterprise systems engineering), Sanmi Koyejo (computer science), and R. Srikant (electrical and computer engineering), in addition to Yuguo Chen, a professor of statistics at Illinois. The multi-disciplinary effort will encourage interaction between scientists in the fields of electrical engineering, mathematics, and theoretical computer science, among others. The institute will need to bring together experts from a variety of disciplines in order get the best collaboration and form new dialogue around data science and dynamical systems.

“The interdisciplinary expertise of our team, as well as the extensive experience in optimization, machine learning, control, statistical simulation and inference, and related topics, make us uniquely qualified to execute our research and education agenda,” said Raginsky. “We are certain that the work supported by the institute will inspire further research both at Illinois and across the Midwest.”

The first phase of research and educational activities the institute focus on four themes:

1. Data modeling and dynamical systems
2. Sampling, interference, and dynamical systems
3. Algorithm design and dynamical systems
4. Decision-making and dynamical systems

The last two themes are of specific interest to He.

“The institute is not only about learning about dynamical systems in data-intensive applications, but also how can we leverage dynamical system perspectives to better understand data science and decision science,” she said. “The rich theory of dynamical systems

provides a natural ground to form a principled understanding of lots of decision-making procedures used in data science, such as their stability, robustness, and implicit bias. There are enormous opportunities to design intelligent systems that could benefit the broader communities.”

All four themes will guide the research directions of the institute, as well as help member faculty plan their outreach efforts. A workshop set for Fall 2020 will kick off an annual series that aims to position Illinois as a leader in the area of data science and dynamical systems. In addition, the institute will sponsor a bi-weekly seminar series to be held in CSL and plans to integrate education activities through workshops, summer schools, seminars, and curriculum development.

This project is funded by the National Science Foundation through the HDR TRIPODS (Harnessing the Data Revolution: Transdisciplinary Research in Principles of Data Science) program, for three years at the amount of $1.5 million. Other University of Illinois faculty involved with the project include CSL’s Yuliy Baryshnikov and Alex Schwing, along with faculty from computer science (Nan Jiang and Matus Telgarsky) and mathematics (Lee DeVille and Kay Kirkpatrick).