By Annabelle “Belle” Hsieh, AI Researcher (intern), Intelligent Human Perception Lab, USC ICT
I am an undergraduate at the University of Pennsylvania, pursuing a dual degree in Computer Science and Systems Engineering, with minors in Mathematics and Statistics. My academic journey began with robotics in high school, where I first saw how code could shape physical systems. Since then, I have been drawn to the breadth of computer science—its ability to influence fields as varied as finance, healthcare, and the arts—and to the challenge of using artificial intelligence to advance both technology and social good.
Research at ICT
This past summer, I worked as an intern at the Institute for Creative Technologies (ICT) at the University of Southern California, within the Intelligent Human Perception Lab led by Professor Mohammad Soleymani. My project involved upgrading LibreFace, an open-source toolkit for deep facial and emotional expression analysis. I developed a new gaze estimation task, designing and testing models to predict yaw and pitch values of gaze vectors from video input.
I experimented with methods including retina center detection, fine-tuning of pretrained masked autoencoders, and the design of a lightweight multilayer perceptron that focused on eye-region features. The final model achieved a mean absolute error of around 13.5 degrees, comparable with other state-of-the-art open-source systems.
Adding gaze estimation to LibreFace strengthened the toolkit’s capacity for analyzing human attention and engagement, and it underscored for me how incremental technical contributions can enhance research resources used far beyond one lab. Working within ICT gave me insight not only into the technical challenges of modeling perception, but also into the collaborative, interdisciplinary environment that sustains meaningful research.
Previous Work
The ICT internship built on prior research experiences. At USC’s Cyber-Physical Systems Group, where I was a SHINE Scholar, I led a project using deep learning to predict RNA mutations in COVID-19 sequences. At Penn, I contribute to the Aerial Robotics team, where I have helped in drone payload recognition and worked with ROS2 and PX4 Autopilot systems to support autonomous flight.
Defining Success
I measure success less by performance benchmarks alone than by whether my work makes technology more accessible and relevant to real human needs. AI has the power to transform, but also the potential to exclude. My goal is to contribute to systems that lower barriers rather than raise them—whether in healthcare, education, or everyday tools.
This orientation has led me to consider new possibilities for my future. When I began at Penn, I imagined a career in software engineering or financial technology. After my work at ICT, I am now also considering graduate study and the pursuit of a Ph.D., to continue research at the boundary of technical rigor and human-centered application.
Collaboration and Mentorship
What made my ICT experience distinctive was the people. I was fortunate to learn from Professor Soleymani’s guidance and from the generosity of PhD students such as Ashutosh Chaubey and Xulang Guan, who shared both technical expertise and practical advice about the research process. Informal conversations—over lunch walks to HomeState or coffee in the cafeteria—often became extensions of the lab, deepening my understanding of both the work and the culture of inquiry that supports it.
Looking Ahead
As I continue my studies at Penn, my focus is on refining the skills—algorithmic, statistical, and engineering—that underpin rigorous AI research. At the same time, I want to remain attentive to the broader implications of this work: who benefits from it, how it is applied, and what values it reflects.
The projects I have undertaken so far are stepping stones. They have taught me the mechanics of model design, the realities of experimental evaluation, and the importance of collaboration. But they have also reinforced a conviction: that the future of AI depends not only on technical sophistication but also on a commitment to accessibility and impact.
I am at the beginning of my career, with much still to learn. Yet the foundation I am building—through research at ICT, through projects at Penn, and through collaborations with peers and mentors—has confirmed for me the path I want to pursue. I hope to contribute to artificial intelligence in ways that are not only innovative but also meaningful, shaping systems that understand and support the people who use them.
//