
Headshot
Author: Bryan Williams | Major: Computer Science | Semester: Spring 2025
During the past spring semester, I began the search for a research topic to pursue under my honors research grant. With so many students gravitating toward artificial intelligence, I decided to take a slightly different path—one that was still technical but offered a unique niche. After looking through the research interests of various professors, I was drawn to Dr. Le’s work in computer vision. It didn’t take long for me to realize that this was the direction I wanted to pursue.
When I started, I had little to no background in computer vision or how it was applied in real-world scenarios. To be honest, it felt overwhelming. I worried I might be in over my head. But thanks to the support and guidance of Dr. Le and her graduate student, Thinh Phan, I was able to get my footing and gradually become more confident in my abilities and understanding of the topic. The main goal of my research was to learn how computer vision models are built and how they function. To help me develop a foundational understanding, Dr. Le started me off with organizing and preparing data—a critical first step in training any computer vision model. That seemingly simple task eventually evolved into a larger project: developing a model capable of monitoring and detecting the behavior of chickens in an enclosed environment.
Through this, I learned how computer vision models rely on neural networks, which consist of an input layer, hidden layers, and an output layer. My task was to create a model that, when given a video feed of chickens in real time, could predict their behavior—whether they were running, eating, sitting, or something else. A well-functioning model would output a distribution of behaviors, with the correct behavior being the most probable. But I don’t just want the model to work—I want it to perform exceptionally well. That means improving the model’s confidence in its predictions, which is critical when using such technology for animal behavior monitoring. A robust, real-time behavior recognition system could have significant applications in animal welfare, enabling early detection of illness or stress.
I began this project during my junior year, which turned out to be my most challenging academic year yet. Balancing coursework and research was no easy task. Weekly meetings with Dr. Le or Thinh held me accountable, but there were definitely weeks when the pressure felt overwhelming. I often found myself staying late on campus, finishing assignments or making progress on the model. This experience taught me a lot about discipline and sacrifice. I came to realize that standing out academically requires more than just showing up—it takes intentional effort and, at times, difficult choices. The key is learning how to make the right sacrifices at the right times.
Throughout this journey, Dr. Le has been an exceptional mentor. She strikes a great balance between offering guidance and encouraging independence. She doesn’t simply give answers but also never leaves me feeling lost.
Despite the challenges, I’m excited about how far I’ve come—and even more excited about where this research could lead. I look forward to continuing my work on the action recognition model and exploring broader topics in machine learning, such as addressing data imbalance. This research journey has been tough, but it’s also been incredibly rewarding.