CAREER: Physics Regularized Machine Learning Theory: Modeling Stochastic Traffic Flow Patterns for Smart Mobility Systems

This Faculty Early Career Development (CAREER) grant will support fundamental research in modeling stochastic traffic flows for smart mobility systems, based on the fusion of classical transportation models and learning techniques. With the goals of mitigating traffic congestions, improving transportation safety, and reducing vehicle emissions, many smart mobility applications require accurate, reliable, and timely traffic information as input.

To meet such needs, this project will lay the foundation of machine learning and traffic flow theory to yield better estimations and predictions of mobility patterns. The method uses transportation domain knowledge to regularize the training process of machine learning. The results will significantly enhance the effectiveness and robustness of those smart mobility applications at both small and large scales.

The research activities can be closely integrated with a set of education and outreach activities that include (i) developing a virtual computing lab to facilitate student educations, researcher engagement, government employee training, and industry collaboration, (ii) modernizing the transportation curriculum with research outcomes, (iii) broadening the participation of k-12 students in the annual summer “Transportation Camps” and underrepresented students in the Artificial Intelligence club of a minority-serving institution. Those activities will help transportation students better recognize the importance of engineering knowledge in the era of smart mobility system.

The goal of this project is to contribute fundamental theories and a set of markedly improved algorithms to traffic flow modeling. Leveraging the concept of physics regularized machine learning, the research could encode both continuous and discretized traffic flow models into Gaussian process for training regularization. This new model can efficiently resolve the common data sparsity and noise issues and facilitate various smart mobility applications.

To accommodate streaming data, this project will also develop a novel physics regularized streaming learning framework that can efficiently improve the model performances in real-time. When dealing with big data, this project can further synergize data of different resolutions, fidelities, and sources to enable sparse Gaussian process and Bayesian committee machine for fast learning.

This foundational research can enormously promote machine learning applications in smart mobility systems and contribute to formulating sustainable, scalable, and robust traffic flow models. This project will bridge the gap between classical transportation methods and data-driven approaches.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.