Just days after successfully defending his dissertation at the University of South Florida鈥檚 Bellini College of Artificial Intelligence, Cybersecurity and Computing, Cole Hill packed his car for a cross-country trip to California. It wasn鈥檛 a celebration or a road trip marking the end of his doctoral journey. It marked the start of a new chapter.
Hill accepted a full-time position with Meta, joining one of the world鈥檚 most prominent technology companies even before walking across the commencement stage.
And walking 鈥� or gait analysis, as it鈥檚 known in the computer vision world 鈥� is what helped Hill get to this point.
Making AI smarter about how people move
Most computer systems recognize people but rely on faces or clothing as the data that fuels AI. But faces can be covered and clothes can change. So, researchers and experts in AI and computer vision are exploring how to use other biometric signals to identify individuals, especially from a distance, using how they move. That鈥檚 gait analysis.
The challenge, Hill said, is that computers get confused when conditions change, such as lighting, camera angles, or new environments. They struggle because they are usually trained on limited data.
鈥淎nd what works on one AI system doesn鈥檛 necessarily translate to another,鈥� he said.
Hill鈥檚 doctoral research focused on that problem. His dissertation, Dynamics-Consistent Representation Learning for Human Motion Analysis and Identification, explored how embedding the physics of movement into AI systems can help them generalize better across environments and conditions.
鈥淢ost AI models are incredibly data-hungry,鈥� Hill said. 鈥淭hey work well on the datasets they were trained on, but what performs perfectly on one system might fail completely on another. My work looks at how we can make AI more consistent by grounding it in the real-world physics of motion, so it understands what鈥檚 happening, not just memorizing patterns.鈥�
Hill said that goal of teaching computers to reason about physical motion grew out of years studying the limits of data-driven systems. 鈥淚f we can help AI understand the mechanics behind what it sees,鈥� he said, 鈥渋t becomes more efficient and far more reliable when the situation changes.鈥�
Hill鈥檚 dissertation committee included USF Bellini College Launch Dean Sudeep Sarkar, Associate Professor Mauricio Pamplona Segundo, Professors Dmytri Goldgof and Kyle Reed and Professor Kevin Bowyer from Notre Dame University. The defense was chaired by Scott McCloskey from Kitware, Inc., Sarkar said the research addresses a key limitation in AI.
鈥淐ole鈥檚 work helps solve one of AI鈥檚 most pressing challenges: its dependence on large, narrowly focused datasets,鈥� Sarkar said. 鈥淏y incorporating physics into training, his models begin with a foundation of knowledge about how the body moves. That makes them more reliable, adaptable and efficient, qualities that are vital for deploying AI in the real world.鈥�
The potential applications of Hill鈥檚 research extend far beyond the lab. His findings could help AI systems interpret motion more accurately and consistently, which could strengthen public safety and defense applications where computers analyze video in complex environments to improve search or even physical rehabilitation technologies. The physics-informed modeling could also support animation, robotics and human-computer interaction, helping machines better predict and respond to human movement.
A doctoral degree wasn鈥檛 in the plan
Hill did not initially plan to pursue a PhD. After earning his undergraduate degrees in electrical engineering and mathematics at USF, he shifted to computer science for his master鈥檚 degree. He met Sarkar and discovered a passion for computer vision.
鈥淚 asked if I could join his research group, then COVID hit and we started working on projects together. I really liked the research thing, so when he said, 鈥業f you want to stay for your PhD, you鈥檙e more than welcome.鈥� I was like, 鈥極h yeah, let鈥檚 do it.鈥�
What began as a master鈥檚 project evolved into a multi-year research partnership that shaped his academic and professional path. He worked with Sarkar and Pamplona Segundo on research that integrates physics-informed modeling with computer vision to improve the reliability of gait recognition for the DoD IARPA/Kitware-funded project. Together, the team also created SynthGait, a large-scale synthetic 3D dataset that generates lifelike human motion under different clothing, viewpoints and walking styles.
Hill also completed internships with the U.S. Department of Defense and the National Security Agency, experiences that strengthened his interest in applying AI responsibly.
鈥淭hose projects emphasized the importance of building systems that perform reliably in unpredictable conditions,鈥� he said.
The gait analysis path led to Meta
Now beginning his career with Meta, Hill doesn鈥檛 yet know what his specific assignments are going to be.
鈥淚 have my brand-new employee ID but I鈥檓 still not exactly sure what I鈥檒l be working on yet. I know it鈥檚 on the AI side of things, but I don鈥檛 have the full details.鈥�
He acknowledges his USF experiences set him up for this role and said he looks forward to coming back to Tampa to attend the graduation ceremonies in December.
鈥淯SF gave me the opportunity to work alongside world-class researchers who genuinely want students to succeed,鈥� he said. 鈥淚t set the stage for everything that is coming next.鈥�