As he entered his final year at Haverford, Ben Bergerson ’24 found himself increasingly interested in the potential of machine learning and neural networks. The physics major turned to the emerging technologies to compare two types of spiral galaxies in our universe for his thesis, an experience he says altered his academic trajectory as he now pursues a master’s in engineering.
Bergerson turned to Professor of Physics and Astronomy Karen Masters for his thesis given her extensive research and connection to Mike Walmsley, a Dunlap Fellow at the University of Toronto who created a new tool specifically for astrophysics research called Zoobot. The tool, Bergerson says, is a neural network that holds significant potential for projects of all scales, from undergraduate research to broader explorations of galactic history.
“Mike’s theory was that he could train his model on a large collection of galaxy images and then individual researchers, like me, could fine-tune this model to a specific task,” says Bergerson. Zoobot played a central role in Bergerson’s thesis, “Classification of Flocculent and Grand Design Spiral Galaxies using Supervised Machine Learning,” which explores two types of spiral galaxies differentiated by the structure and definition of their radiating arms.
There are not many existing datasets in which flocculant and grand design galaxies are classified, Bergerson says, and his thesis relied on one containing about 500. Using Zoobot, he developed a model that allowed him to classify more than 30,000. Bergerson’s dataset, he says, could be analyzed further using more traditional methods to interpret patterns within spiral galaxies that, in turn, can help us better understand our universe.
Bergerson is enrolled in Haverford’s Accelerated Masters Program with the University of Pennsylvania, a first-of-its-kind partnership between a liberal arts college and an Ivy League engineering program. He says his thesis at Haverford was the perfect introduction to machine learning concepts that will continue to influence his career path.
“I plan to continue exploring deep learning and neural networks. I don’t know if I would be doing this if it weren’t for my thesis,” he says. “I’m not sure where my next degree will lead me, but a career in machine learning or data science is quite possible.”
But wherever Bergerson winds up, he will be hard-pressed to fund a community as supportive and influential as he did in Haverford’s physics department.
“I have had a wonderful four years at Haverford College, and I accredit much of this to the amazing physics department. I was so lucky to have such close relationships with so many brilliant physicists,” says Bergerson. “The care that they have for the students has created a culture within the physics department that seems very unique to Haverford, and I am lucky to be a part of this.”