2021-2022 Frontera Fellows Cohort - So Long and Good Luck!

TACC's Frontera Computational Science Fellowship for 2021-2022 came to a close at the end of May 2022. The program provides a year-long opportunity for talented graduate students to compute on Frontera, the most powerful academic supercomputer in the world, and collaborate with experts at TACC.

TACC caught up with the Fellows to hear about their experiences, lessons learned, and why they would recommend other students apply to the Fellowship.

Liliana Cabrera Gallegos

School: Colorado State University

Field of Research: PhD candidate, Computational Organic Chemistry

Can you talk about the project(s) you worked on during your time as a Fellow?

My project was based on an idea to assist experimental organic chemists. I used computational tools such as machine learning and mechanistic studies using quantum mechanical calculations to predict the best optimal reaction conditions and improve reaction outcomes. In a perfect world, you go into a lab, perform an experiment, and get a high yield of the desired product. However, the reality is that you go into a lab and get 10% of your desired outcome, so you must optimize the reaction. My project was based on finding a model that can predict these reaction conditions based on a mechanistic understanding of the reaction.

What did you learn from a scientific/computational point of view, or from a professional perspective?

Before the Frontera Fellowship, I didn't have much experience with machine learning. I gained a lot of confidence in this area during my time as a Fellow. I learned a lot through hearing from the other Fellows about their struggles with machine learning. I was also able to take advantage of the resources the program offered.

What did you enjoy most during your time as a Fellow?

I enjoyed talking with the other Fellows. Our research is so different, but I enjoyed hearing about the many applications where machine learning can be used. I also enjoyed the bi-weekly talks with our mentors.

How will this Fellowship help you moving forward in your field of research?

The knowledge I've gained by utilizing HPC has given me the confidence to apply machine learning to future projects. I have a year left before I graduate and plan on taking a postdoc position. Before the Fellowship, I wouldn't have applied for certain positions because I didn't feel qualified.

What would be your advice for incoming Fellows, and why should others apply to the program?

I would advise prospective students to write a proposal based on what inspires you. You've got nothing to lose, so take the risk and apply for the Fellowship. Because I went through this program, I get to talk about the research I performed here at TACC, and I'm excited to share my story.

Gurpreet Singh Hora

School: Columbia University

Field of Research: PhD candidate, Civil Engineering and Engineering Mechanics

Can you talk about the project(s) you worked on during your time as a Fellow?

I worked on a project to reconstruct natural disasters using data slices. Tracking and predicting the fate of air pollutants, like toxic chemicals and smoke, is a priority for national emergency agencies. Typically, satellite images are used to monitor airborne dispersion, but they are 2D and cannot provide information on the 3D structure of the process. I'm also still working toward the development of a 3D rapid response framework for airborne dispersion in the atmosphere. Eventually, the technology resulting from this project will provide an accurate and efficient framework that can be used to quickly predict the fate of pollutant plumes and their interaction with 3D environments.

What did you learn from a scientific/computational point of view, or from a professional perspective?

The Fellowship provided me with a unique opportunity to tackle an important and unexplored topic at the interface of two well-known disciplines: turbulent flows and machine learning. The generous computing resources supported the generation of an extensive database of high-fidelity turbulent flow fields, which were used for the training and validation of the machine learning algorithms. The research I've performed has advanced my fundamental understanding of turbulent flows and machine learning algorithms and enabled me to push the boundaries when it comes to approximating non-linear, multiscale, and chaotic flow phenomena.

What did you enjoy most during your time as a Fellow?

I've enjoyed the continuous support from the mentors and the top-of-the-line research and computing resources that were made available. This has been a fantastic opportunity for personal and career growth, and my time as a Fellow has contributed to my goal of becoming a well-established independent researcher.

How will this Fellowship help you moving forward in your field of research?

I was trained on state-of-the-art computational tools, received support for the development of a high-fidelity computational fluid dynamics dataset, and received training and validation of Deep Learning algorithms. These opportunities enabled me to make progress in my research. My interactions with mentors also spurred new ideas and opened fruitful avenues for future research. The knowledge I gained in computational science — along with the preliminary results that were generated as part of this effort — bolstered my confidence and motivated me to expand my research program to more realistic scenarios.

What would be your advice for incoming Fellows, and why should others apply to the program?

I would advise students to utilize the resources, attend as many meetings as possible, and take full advantage of interactions with mentors. Students can acquire knowledge on state-of-the-art computing tools from the skilled mentors at TACC. As a Fellow, you will have access to a powerful, HPC ecosystem, which can greatly accelerate your research. The financial support and tuition allowance will provide much-need support for the continuation of your studies.

Yui Tik "Andrew" Pang

School: Georgia Institute of Technology

Field of Research: PhD candidate, Biophysics

Can you talk about the project(s) you worked on during your time as a Fellow?

My project focused on the use of artificial intelligence and deep learning methods to help simulate long timescale protein dynamics on largescale supercomputers. You may have heard of "AlphaGo," a program developed by Google that taught a computer to play chess. Eventually, AlphaGo was able to beat the best human player in the world. Incorporating other key developments in AI, our project is attempting to do something similar, but instead of playing chess, it guesses how a protein transforms itself from structure A to structure B.

What did you learn from a scientific/computational point of view, or from a professional perspective?

This was my first time working on a deep learning project. I learned how to write custom loss functions and training loops with TensorFlow and how to compile scientific packages on a POWER9 supercomputer. I didn't know what to expect going into this Fellowship, but I learned a lot more than I expected. I'm grateful for both the financial and technological support.

What did you enjoy most during your time as a Fellow?

I was able to present my Frontera project at the annual Biophysical Society Meeting in San Francisco. Having the opportunity to address the hundreds of people who were in attendance was an amazing experience, and our cohort received great feedback from experts and potential Frontera users.

How will this Fellowship help you moving forward in your field of research?

My project will have a real impact on what can be done with molecular dynamics simulations. I have other studies that can also benefit from the AI model from this project, including one related to the SARS-CoV-2 spike and one about protein folding mechanisms. I look forward to seeing projects from other molecular dynamics researchers making use of this model. We are still working on the manuscript; hopefully, we can publish our work this year!

What would be your advice for incoming Fellows, and why should others apply to the program?

This is a unique opportunity to work alongside a team of experts, and you get to use some of the best supercomputers in the world. When I ran into any compiling issues or had difficulties understanding some of the deep learning concepts, the mentors were always ready to help. Reach out to your mentors and the others in your cohort and learn as much as you can. The more work you put into your time as a Fellow, the more rewarding it will be. You also receive a nice financial stipend, which helps with your expenses.

Hongyuan Zhang

School: University of Minnesota Twin Cities

Field of Research: PhD candidate, Mechanical Engineering

Can you talk about the project(s) you worked on during your time as a Fellow?

In situ adaptive tabulation (ISAT) is a method to accelerate computation by building a table during the simulation. I'm working on developing a version of ISAT that allows parallel computation. If I run a simulation using 100 processes, each process needs to build its own table. Since many records will be redundant, this is a waste of memory. My work is to use the shared-memory model to share the table among all processes in a computational node to save computation and memory resources.

What did you learn from a scientific/computational point of view, or from a professional perspective?

I learned a lot about coding knowledge with regard to the Message Passing Interface (MPI) shared-memory model. Before my cohort started, I expected to learn mature-model and user-level MPI processes. After I started, I realized that the MPI shared-memory model was still quite new and could only support simple functions. For my research, this would be insufficient, and I knew that I needed to bridge the gap. That's why, during my time as a Fellow, I not only learned mature-model and user-level MPI processes, I also learned techniques like memory management and inter-process communication.

What did you enjoy most during your time as a Fellow?

I'm the type of person who is interested in learning new techniques. During my year as a Fellow, I spent a lot of time on using computational resources to test my ideas on MPI. I really enjoyed learning new things and developing code.

How will this Fellowship help you moving forward in your field of research?

My previous work in Computational Fluid Dynamics (CFD) mainly focused on developing new physical models to capture detailed physics. Many CFD researchers also focus on algorithm optimization and acceleration, which is one of my weaknesses. This Fellowship has helped to bolster my skills in these areas, which will allow me to conduct more research in the future.

What would be your advice for incoming Fellows, and why should others apply to the program?

Those who want to develop novel HPC methods should apply to be a Fellow. This program supports research using all computational methods and has huge resources, which are critical for developing new methods. In addition, the people here can help you with research and applying machine learning to various fields.

Learn more about the Frontera Fellowship program.