REU FoDOMMaT Students Dive Deep into Machine Learning and Deep Learning

Each summer, NCSA brings undergraduate students from across the country to work on cutting-edge research projects in an enriching environment. The Research Experience for Undergraduates (REU) FoDOMMaT program at NCSA is a 10-week on-site program where students develop machine learning (ML) and deep learning (DL) programming skills to solve real-world problems. REU FoDOMMaT is one of many REU programs funded by the U.S. National Science Foundation nationwide.

FoDOMMaT, or The Future of Discovery: Training Students to Build and Apply Open Source Machine Learning Models and Tools, pairs students with two research mentors. One mentor leads the project’s research, while the other offers expertise in machine learning. Mentors mostly hail from outside the AI space, from civil engineering, sociology or other fields, says Priyam Mazumdar. Mazumdar is a doctoral electrical and computer engineering student at the Grainger College of Engineering and the REU graduate research leader for the FoDOMMaT program. He co-developed ML training for this program this year’s FoDOMMaT programming with NCSA’s REU FoDOMMaT principal investigator (PI), Volodymyr Kindratenko.

“REUs are experiences where an undergrad student is assigned a professor or researcher to work with on whatever problem they’re currently trying to solve,” Mazumdar said. “Our best practices were developed to allow students to explore a wide range of problems, unlike other REU programs where students may only work on niche problems.”

Mazumdar’s approach to designing the FoDOMMaT ML training is unique because the students are introduced to a new problem each week. During the first week of the ten-week program, students get a crash course in AI and ML, assuming they have basic programming skills and general math knowledge, building students up from the ground up. It’s five grueling days of four to five hours a day dedicated to learning the basics of AI.

“I give them everything I can about AI, the very basics of machine learning, linear regression and building up the core concepts and ideas that come from simple machine-learning models. We move them up to slightly more complicated concepts like support vector machines, random forest, then deep-learning frameworks such as PyTorch to build larger neural network-based architectures.”

In the next two to three weeks, students take a lot of time to understand the data they’re working on with discussions on exploratory data analysis.

“It’s kind of like putting a LEGO set together. They are doing a full-scale analysis of the data just to see what they are working with. And while doing that, they pick up the tools and techniques they need to learn for their specific tasks,” Mazumdar said.

What’s truly unique about the REU FoDOMMaT program is the weekly reading group. On Fridays, students receive a paper from which they must extract the key concepts and try to solve a problem. Mazumdar described reading a paper and figuring out more or less what’s going on at a glance as an essential skill for future researchers.

Dr. Kindratenko and I agreed that reading a paper is one thing, but actually building on that paper is a completely separate problem altogether.

Priyam Mazumdar,  REU graduate research leader for the FoDOMMaT program

Once students have skimmed the article to understand as much as possible, they work on implementing the full pipeline to solve it.

“Just because I read that paper doesn’t necessarily mean I could build my own GPT. That is the knowledge gap that I felt was there,” Mazumdar said. “Each week, we’ll implement some state-of-the-art architecture. I put students in two groups, and each group had to present to me which architecture they wanted to work on and present one novel change they wanted to make to try to improve it.”

Getting into the REU FoDOMMaT program has become quite competitive, said Olena Kindratenko, senior research coordinator at NCSA, who administers the REU FoDOMMaT program. Students apply in late fall and describe what motivates them, why they want to do research and must submit their unofficial transcripts.

Sophia Witola Reyes is a rising junior at the University of Illinois at Urbana-Champaign and in the REU FoDoMMaT program. She’s been working on a project called Visible Nutrition: Applying Artificial Intelligence to Achieve Personalized Precision-Based Nutritional Monitoring & Guidance, which uses AI to optimize approaches to dietary assessment and guidance.

“Coming into the program, I had little to no experience with software development or machine and deep learning, so that, along with the prestige of NCSA, made me very anxious about entering the program,” Witola Reyes said. I was determined, though, to make sure that this didn’t prevent me from coming in with an open mindset and readiness to learn.”
Witola Reyes says the experience this summer has allowed her to grow as a researcher and programmer.

“My biggest takeaway from the program’s training is how applicable ML/DL methodology and technologies are to literally anything. In our training, I really appreciated how, before getting into the implementation of a model, Priyam would show us ways it is currently being applied in real life,” Witola Reyes said. “Moreover, we got more insight into how NCSA is currently doing that through its numerous projects. I’m excited to find ways the concepts I learn in my BioE classes could be approached from a machine-learning lens.”

Witola Reyes has grown to see herself as a researcher. “I’ve also gotten a lot of insight into what it means to be a researcher: working independently, asking questions, holding yourself accountable in terms of setting and reaching goals, and being a strong communicator,” she said.

Volodymyr Kindratenko is the director of the Center for Artificial Intelligence Innovation and is the PI of the REU FoDoMMaT program. He’s also Witola Reyes’s mentor. He says that the program provides opportunities in STEM research to students, especially those from underrepresented groups, to gain experience in ML and deep learning, fields he says are in high demand nationwide and worldwide.

“The REU FoDoMMaT program is not just about the technical knowledge gained; it’s about fostering a community of learners who are passionate about discovery and innovation,” Kindratenko said. “Ultimately, the hope is that students will leave the program with a deeper understanding of the subject matter, improved research skills and a sense of empowerment to pursue further education or careers in STEM fields.”

Dan Katz is the chief scientist at NCSA and co-PI of the REU FoDOMMaT program. He believes the program gives students opportunities to explore alternate career paths they might not have otherwise. He hosted a networking event with FoDOMMaT and Students Pushing Innovation (SPIN) students to share his career journey from industry, academia, national laboratories and government.

“I wanted to share this with the students because my career has been successful even without taking a traditional path. This included talking about where I had choices, what I chose, and how I made those choices. My intent was to show students that they don’t have to follow any one path,” Katz said.

Angela Lyons is an Agricultural & Consumer Economics professor at the U. of I. She and research scientist Aiman Soliman with NCSA’s Data Analytics and Center for AI Innovation mentored two students, Anushka Mazumdar and Cassi Chen, in the REU FoDoMMaT program.

“I appreciate the opportunity to work with talented young data scientists from around the country. Mentoring students such as Anushka Mazumdar and Cassi Chen allows me to help shape the next generation of data scientists, influencing their career paths and future successes. It is profoundly rewarding to watch their rapid growth in understanding and applying data science within just a few short weeks,” Lyons said.

“This summer, Anushka and Cassi have been working on a unique project, A Geospatial Analysis of Syrian Refugees in Lebanon Using Machine Learning, which bridges social science with data science for social good. Seeing them use cutting-edge machine learning and geospatial analysis to aid humanitarian and development organizations such as the United Nations High Commissioner for Refugees and the World Food Programme in better tracking and targeting refugee populations is very fulfilling,” Lyons said. “Over the years, as we have worked on various aspects of this project, the students continually bring fresh perspectives and innovative ideas that our team has not considered.”

Of course, doing this research comes with obstacles along the way. Anushka Mazumdar will be a junior in electrical and computer engineering this fall at The University of Texas at Austin.

“Some challenges I have faced are the sheer amount of work it takes to create a dataset from scratch. A lot of our data was on maps and Tableau format, so we had to scrape it by hand,” she said. “We also struggled with whether or not to prioritize understandability or accuracy during our data preprocessing steps, but with many trials and errors, as well as advice from our mentors, we were able to determine the best way to approach it.”

Despite these challenges, Anushka Mazumdar greatly benefited from this summer’s REU FoDOMMaT program. “From this experience, I’ve learned about working on a team long-term and what it’s like to coordinate tasks so thoroughly. I’ve learned a lot technically, but the working relationships and friends I’ve made along the way will definitely last beyond this program,” she said.

---

About NCSA

The National Center for Supercomputing Applications at the University of Illinois Urbana-Champaign provides supercomputing, expertise and advanced digital resources for the nation’s science enterprise. At NCSA, University of Illinois faculty, staff, students and collaborators from around the globe use innovative resources to address research challenges for the benefit of science and society. NCSA has been assisting many of the world’s industry giants for over 35 years by bringing industry, researchers and students together to solve grand challenges at rapid speed and scale.

---

NCSA SUMMER PROGRAM RECOGNITIONS

2024 Summer REU FoDOMMaT Students

Cassi Chen
Hao Huang
Anton Matchev
Anushka Mazumdar
Annapoorna Narayan
Nina Trousdale
Erik Velazquez
Sophia Witola Reyes
Adam Yang

2024 SPIN Students

Arrhan Bhatia
Kai Chandu Karadi
Haoqi Chen
Yash Ejjagiri
Wenbo Huang
Hunter Jiang
Vira Kasprova
Abby Lachman
Qi Long
Sarn Mukhopadhyay
Zaeem Qureshi
Dhruv Srivastava
Eric Yu
Ray Xu
Ziqi (Zoe) Xu
Rui Zhou

Thanks to our Outstanding REU FoDoMMaT Mentors

Bruno Abreu
Kastan Day
Eliu Huert
Volodymyr Kindratenko
Sheng Wang

Thanks also to this summer’s Outstanding SPIN mentors

Mohamad Alipour
Jim Basney
Phuong Cao
Roland Haas
Volodymyr Kindratenko
Bertram Ludäscher
Xin Liu
Jill Naiman
Santiago Nunez Corrales
Anastasia Stoops
Yuxiong Wang

REU FoDOMMaT and SPIN students participated in the 2024 STEM Career Exploration on July 26, which allowed them to highlight their research results to other REU students, high school and community college students, faculty and science teachers. They were among the 190 student presenters, with around 560 attending the event.

More on training work carried out can be found in an IEEE article titled, Training Next-Generation Artificial Intelligence Users and Developers at NCSA.