Renowned Landscape Designer Hood '81 Wins Vincent Scully Prize
10/02/2024 in Alumni, College of Agriculture and Environmental Sciences
By Jamie Crockett / 03/12/2024 College of Science and Technology
EAST GREENSBORO, N.C. (March 12, 2024) – A research team in the College of Science and Technology at North Carolina Agricultural and Technical State University has conducted a pilot project using models available through Microsoft’s Azure Open AI Service to develop a system that will lead to advancements in traffic management using physic informed deep learning. The pilot, led by principal investigator Leila Hashemi-Beni, Ph.D., and applied science and technology doctoral student Tewodros Syum Gebre, received support from Microsoft under the company’s “Accelerating Foundation Models Research” (AFMR) collaboration.
The AFMR program provides academic researchers with access to state-of-the-art foundation models through Azure AI Services, with the goal of fostering a global AI research community and creating robust, trustworthy models that help further research in disciplines ranging from scientific discovery and education to healthcare, multicultural empowerment, legal work and design. The initiative’s program includes 200 projects at universities in 15 countries, spanning a broad range of focus areas.
As one of the awardee teams, N.C. A&T researchers wanted to use their expertise to enhance current monitoring systems because traffic congestion contributes to an increase in emissions. The Environmental Protection Agency references the Inventory of U.S. Greenhouse Gas Emissions and Sinks 1990-2021 report confirming “transportation accounted for the largest portion (29%) of total anthropogenic U.S. greenhouse gas emissions in 2021.” The report underscored light-duty vehicles made up approximately 58% of that portion, while medium- and heavy-duty trucks were responsible for 23% of emissions.
“Currently, traffic management requires a lot of manual effort, and that can be quite limiting,” said Hashemi-Beni, an associate professor in the Department of Built Environment with research interests in remote sensing and geospatial data science and their applications including transportation planning.
“Tewodros has done very advanced work related to data-driven methods for transportation for his research dissertation, but now he has taken an additional step by combining deep learning with a complex mathematical model that is already in use in transportation through the state-of-the-art foundation models,” she said. “He has not limited himself and essentially brought the advantages of two different modeling systems to the table which is called physic-informed deep learning. He has contributed a lot to the proposal and project and I’m so proud of him.”
“I want to combine what I’ve learned in civil engineering and data science to solve transportation issues. The goal is to enhance the well-being and safety of people.” said Gebre. “For example, if someone is stuck at a red light when there is no vehicle traveling the other way, they are adding to the amount of emissions just waiting on the light to change.”
While emission contributions to climate change is a factor the team considered while conducting this project, Gebre also emphasized the immediate loss of life he hopes can be avoided.
“Imagine someone waiting at a red light for a long time and deciding to go through it. This could lead to an accident if another driver shows up.” he said. “We should handle transportation in a smarter and more efficient way to save lives and take care of the environment.”
According to a recent World Health Organization report, approximately 1.19 million people die each year in road traffic crashes.
The research team received access from Microsoft to use Generative Pre-trained Transformer 4 (GPT-4), an advanced OpenAI model “allowing users to interact with data using conversational language.” The researchers built a model that could help them learn the traffic pattern and estimate how many vehicles can be expected on a particular stretch of road during a specific timeframe. So, in case of GPS or any traffic sensor failure or if a person’s data is interrupted, these models should be able to provide reasonable predictions.
“On top of that, we wanted to make sure that people can interact with the traffic prediction model using natural language,” said Gebre.
Thus, the team integrated AI with the traffic models, which would allow users to ask questions about traffic flow and the tool developed with GPT-4 will respond to them. The pilot project laid the groundwork for developing the methodology for this novel concept that at some point will be implemented publicly.
Enhanced traffic management will be beneficial for additional stakeholders including departments of transportation, city planners and law enforcement. This research is built upon Hashemi-Beni and Gebre’s work with the NC Transportation Center of Excellence on Connected and Autonomous Vehicle Technology (NC-CAV).
Media Contact Information: jicrockett@ncat.edu