About 20 UC Merced students spent the past two weeks working at Lawrence Livermore National Laboratory (LLNL) to see if they can solve a problem that could have a significant impact on cardiology.
The annual Data Science Challenge (DSC), a two-week, full-time internship at LLNL, this year teamed students from Merced and UC Riverside. They attempted to see if machine learning could address a gap in the information provided by the common electrocardiogram (ECG) test.
ECGs are a noninvasive, cost-effective way to diagnose many heart conditions. The tests map electrical activity in the heart so that people trained to read the graphs can detect deviations from normal. The problem is, they don't provide enough detail for many applications.
At this year's DSC, undergraduate students, their graduate student team leaders and mentors from LLNL explored data-driven approaches to reconstructing electro-anatomical maps of the heart, combining input from the standard 12-lead ECG with advanced machine learning techniques. This year's challenge problem was developed by Mikel Landajuela, senior staff scientist in machine learning.
“As a team, we had multiple brainstorming sessions on ways to solve the main problem. Also, for the easier problems I learned and made a deep-forest algorithm for training and compared it with a teammate’s methods of solving it,” said Brandon Pardi, a senior computer science and engineering (CSE) major. “For the main, most difficult, challenge, I implemented a skip connections architecture model with the potential for a very high yield, and after that challenge was completed, it was just about seeing how high we could go.”
Javier Miranda, another senior CSE major, worked on modeling.
“I was investigating various machine learning models that we could implement in our classification tasks to improve accuracy and runtime efficiency, and I created tables and graphs to visualize our findings,” he said. “I also compared and evaluated the classification models we used and studied their complexity. I also helped set up our development environment.”
Neither student had participated in the DSC before, but both said it was an invaluable opportunity.
“I've learned so much,” Pardi said. “The first few days were a struggle with imposter syndrome, but after a couple days of discussions, making contributions and reminding myself that among all the people who applied I got accepted, it got easier, and with time here I have felt more confident in my abilities. As with all programming challenges, it was frustrating at times, but I had a good team to work with, and while we all have varying levels of machine learning experience, everyone on the team — and in the whole challenge for that matter — is really intelligent.”
“Every day, I learned and grew immensely. This program pushed me in numerous ways, especially because I had no prior experience in data science and machine learning,” Miranda said. “Now, I feel confident in applying my newly acquired skills to tackle real-world problems using data.”
The DSC opens a huge doorway for students who might not have ever thought about working at a national lab or using their math skills in other fields as career possibilities.
Pardi said he already knows he will be looking for an internship at LLNL after he graduates.
“As much as possible, we want to remove barriers and set up students with the skills and training employers want, and give them as many options as possible,” said Professor Suzanne Sindi, chair of the Department of Applied Mathematics and the Merced faculty lead for the DSC.
LLNL computer scientist Brian Gallagher, who serves as DSC director alongside LLNL data scientist Cindy Gonzales, said the two campuses have previously not worked together at the same space at the same time. For the past two years the event was held virtually, but this year he and the other organizers thought it would be a good opportunity for the students from both campuses to team up.
“It worked out great,” he said. “The students work together really well, and they all got along and mingled so much that I couldn't tell who was from Merced and who was from Riverside.”
Having twice as many people at the DSC this year created a little more work, but the students also had a new space to work in. Just outside the gates of LLNL, a renovated building has been named the University of California Livermore Collaboration Center after the long-standing friendship between the UC and LLNL.
“This space makes it easier for us to have more partnerships with the lab,” Sindi said.
During their two-week stay in Livermore, the students toured some labs, attended professional development sessions, worked on the data science problem, met with and heard lectures by mentors, had social events and, at the end, presented the results of their work at a scientific poster session.
UC Merced graduate student Maia Powell, a past DSC participant, designed this year’s T-shirts.
“During the year I participated, the pandemic had just started, but the Data Science Challenge allowed for an opportunity to tackle an interesting chemistry problem,” she said. “The Data Science Challenge serves as an opportunity to get hands-on data skills, collaborate across disciplines and network with individuals at the lab.”
This year's mentors included Lorenzo Booth, a Ph.D. student in Electrical Engineering and Computer Science, and Asees Kaur, Ph.D. student in Applied Mathematics. Neither one had participated before.
Booth said he applied for the DSC because his labmate told him what a positive experience it was.
“Things went great. It certainly kept me fully busy the entire day,” Booth said. “My team was a good mix of people at different points in their development as data scientists — we all learned from each other.”
Touring LLNL’s National Ignition Facility (NIF) was a special opportunity for all the participants. The NIF allows researchers to study materials at extreme pressures, temperatures and densities. The facility achieves temperatures and densities more than an order of magnitude greater than those in the sun’s core and pressures that far exceed those at the core of Jupiter, according to the lab.
That kind of power impressed everyone.
But the participants said they were also impressed with the staff and researchers at the lab, and the interactions they had with the researchers and each other.
“The facilitators and researchers who came to visit us during the lunch sessions were very friendly and open to answering all sorts of questions,” Booth said.
Kaur agreed.
“The thing that got me going was the coordination and cooperation not just among the team members but also among different teams,” she said. “Our hosts and mentors at the challenge were very supportive, and it was a great learning and networking opportunity for me.”
Both graduate students said they learned a lot about managing diverse teams and they felt the undergraduate students had a chance to see how graduate students approach problems and to learn about their research. They all also learned about the analytic techniques being applied toward large-scale experiments at the national lab.
Outside of all the learning opportunities, one thing stood out to Pardi the most.
“Everyone who participated came from a very unique background and we all had diverse interests,” he said. “So, the most interesting thing about the challenge for me was the people I met and the friends I made.”
