By Josh Baxt
CSE Assistant Professor Rose Yu, who recently arrived from Northeastern University in Boston, is developing physics-guided machine learning techniques to model spatiotemporal data. She investigates traffic flows, human mobility and fluid dynamics, but her passion for computer science began more humbly.
“I think it was because of my love for computer video games,” said Yu. “I played a lot of World of Warcraft in high school.” That pastime sparked an early interest in computers and later in machine learning.
Yu earned her PhD at USC, where she was honored for best dissertation. Before joining the Northeastern faculty, she was a postdoctoral fellow at Caltech.
“My time at Caltech was fascinating because it’s such a small and highly interdisciplinary community,” said Yu. “I learned a lot from people in physics, biology, control and aerospace engineering”
While she enjoyed her time at Northeastern, she had been eager to move back to California to be closer to family. “I received several offers in California, but UC San Diego stood out because of its highly dynamic, innovative and interdisciplinary research environment,” she said.
Yu wants to develop machine learning algorithms that can make sense of large-scale spatiotemporal data in many fields, including self-driving vehicles, subatomic particles, biological molecules, ocean currents and even people. She approaches these problems by integrating physics into machine learning. One research area is traffic modeling.
“Traffic can be more challenging than particle physics,” said Yu. “Traffic flow has multiple dimensions of complexity in terms of the dynamics and the interactions, and it also has an extra layer of complexity from human behavior.”
To better account for this, Yu’s group models traffic by combining physics-based models with machine learning.
“If we want to model vehicle trajectories, there are underlying physical laws governing their motions,” said Yu. “Current machine learning methods don't take these laws into account, which means their predictions often are not physically meaningful. I want to develop faster models and algorithms to analyze space-time data, and I want the predictions from these models to be useful for scientists.”
Her approach has numerous applications in science and engineering: epidemic modeling, self-driving vehicles, drones that fly more stably in high winds. Recently, Yu and colleagues at UC San Diego contributed to the Centers for Disease Control and Prevention’s COVID-19 mortality forecast.
A passionate teacher, Yu takes a hands-on approach, believing the best way for students to learn is to engage in the research process.
“I taught an undergrad class introducing students to computer science research at Northeastern,” said Yu. “It showed how they can conduct research, from doing a literature survey to requesting mentorship to thinking critically about certain problems and how to explain their ideas clearly and write them down. I’d like to continue this type of class at CSE.”
Yu is also a strong advocate for women in machine learning, and has worked to broaden participation from underrepresented groups, extending their opportunities to conduct and present research.
“I want to support more students from minority groups in machine learning,” she said. “If they're interested, I'm always here to help them.”
Yu is one of four new faculty members to join the distinguished community of CSE faculty over the next two academic years.