Mentor: Dorian Abbot, Professor, Geophysical Sciences
Jay Shen is a fourth-year physics and computer science major at the University of Chicago. His background is highly quantitative and research-intensive, with an emphasis on math, physics, and computer science. Currently, Jay is designing deep learning models to accomplish tasks in a drug-design workflow, whether that be property prediction or candidate generation. Jay is also interested in physics for dynamical systems, especially as relevant to topics in physics and geophysics.
As a 2025 Polsky Research Fellow, Jay Shen explored model compression techniques for dynamical system emulators with Professor Dorian Abbot. The assumption is that these models are highly overparameterized, and that this overparameterization is connected to overfitting. Shen worked on pruning this overparameterization to see if model generalizability and accuracy improved.
“I learned about climate emulators and how they work. I also learned about how to train large neural network models on distributed training clusters. My pruned model was shown to improve in accuracy compared to the unpruned model. I plan to continue with this project and hopefully use this experience to apply to graduate schools.”