In the world of artificial intelligence (AI), the integration of physical principles into deep learning is marking a turning point. Rose Yu, associate professor at the University of California, San Diego, is leading this emerging field known as “physics-guided deep learning.” Her work has not only introduced new techniques in neural networks, but has also enabled advances in practical applications such as traffic prediction, climate modeling, and drone stabilization.

The inspiration behind physics-guided deep learning

Yu’s interest in combining physics and AI began with an everyday problem: traffic in Los Angeles. In attempting to predict traffic patterns, Yu and his team modeled vehicle flow as a diffusion process, similar to fluid flow. Using massive data collected from traffic sensors, they managed to extend traffic predictions from 15 minutes to an hour, a significant improvement that was even implemented in Google Maps in 2018.

This innovative approach laid the foundation for applying physical principles in other fields. For example, Yu used fluid dynamics to model turbulence in oceans and atmospheres, speeding up climate simulations by up to 1,000 times in three-dimensional configurations. These advances could improve the prediction of hurricanes and other extreme weather events.

From theory to practice: applications in multiple disciplines

Physics-guided deep learning has applications beyond traffic and weather. In the medical field, Yu has explored how to model turbulence in blood flow to prevent strokes. In aerospace, he has worked on stabilizing drones using neural networks that model the turbulent interactions between propellers and the ground.

Yu is currently collaborating with scientists on nuclear fusion energy projects. His goal is to develop AI models that predict the behavior of plasma, an extremely hot and unstable phase of matter, in real time. This breakthrough could be crucial for controlling fusion reactions and advancing toward sustainable energy sources.

The future: AI as a scientific assistant

Yu envisions a future where AI acts as a scientific assistant, capable of discovering physical principles and generating hypotheses. His team has already developed algorithms that identify fundamental symmetries, such as the constancy of the speed of light, from raw data. These tools could revolutionize scientific research by automating repetitive tasks and allowing researchers to focus on creative aspects.

However, Yu emphasizes that AI will not replace human scientists. Its goal is to complement human work, not replace it, ensuring that creativity and critical judgment remain central to the scientific process.

Continue your professional career

If you are interested in exploring how physics and artificial intelligence can transform industries, consider the Master of Science in Strategic Management with a Specialization in Information Technology. This program will prepare you to lead technological and scientific projects in an ever-evolving world.

Source:

WIRED – Improving Deep Learning With a Little Help From Physics