About

I will be at NeurIPS 2025, a short drive away from UCSD in San Diego, California. Feel free to reach out if you are interesting in talking about generative models for priors in science and world modeling, Diffusion models, and the awesome applications of these topics!

I am a first-year Data Science Ph.D. student at the University of California, San Diego working under Professors Rose Yu and Yian Ma. Before starting my Ph.D., I graduated from with an A.B. in Statistics & Mathematics and an S.M. in Applied Mathematics from Harvard University and wrote my senior thesis under Professor David Alvarez-Melis. I also spent a year in the quantitative trading industry.

My research interests center on improving generative models to meet real-world constraints. Specifically, I am interested how scientific and mathematical insights on the statistics and geometry of data can synergize with generative processes. I also research methods to improve the inference speed of Diffusion and Flow Matching Models. I began my undergraduate research exploring how to make state-action spaces interpretable for safe reinforcement learning in healthcare. Afterwards, I studied the connection of loss landscape geometry and generalization in deep learning through the lenses of parameter space symmetries and model merging.

You can reach me at a5rojas[at]ucsd[dot]edu.