dustin nguyen

This page is no longer maintained as March 2024. I will provide updates when I publish again.

Pre-March 2024:

I am actively working on both machine learning and astrophysical research. In the last two years, I have primarily focused on scientific machine learning with Universal Differential Equations.

The most up-to-date list of my publications can be found on ADS or arXiv.

machine learning research 🤖🧠

2 papers into ML workshops (1 independent, 1 first author)

Neural ODEs/PDEs

Using neural networks as individual terms within ODEs/PDEs to learn physics from data.

Protein Function Prediction

Kaggle CAFA 5 protein prediction competition: As a part of the Erdos Institute Data Science Bootcamp, my team and I studied how to develop models for protein function prediction using frequently occuring protein labels from T5, ESM2, and ProtBERT pre-trained embeddings.

Our project placed within the Top 5 projects, out of 33 teams, of the bootcamp.

astrophysics research 🔭🌌

4 first author papers, 2 co-author papers

The bulk of my thesis research is focused on developing models that explain observations of galactic starburst winds. This is important because galactic winds, or outflows, are closely related to how galaxies evolve. Below is an example of output from a Cholla simulation. These are 2D slices through the 3D volume for several different thermodynamic quantities. The plot showcases highly mass-loaded flows undergoing an instability and becoming unstable to cool filament formation (Nguyen+2023).

One of my favorite papers I wrote is listed below. In this short 5 page paper, we derive analytic solutions from steady-state coupled ODEs, verify them using time-dependent 3D simulations, and then highlight the observational characteristics of the models to be used in comparison to forthcoming data from future space missions.

My 3D time-dependent simulations are done using GPU-accelerated astrophysical code “Cholla” (primarily in C++ for the version I used). Analysis of simulation results is always done using Python. For works ofcused on semi-analaytic models, the steady-state coupled ODEs are solved in either Julia, JAX, or Python.

The papers can be found within my ADS or arXiv.