Comfort zones are so comforting, particularly in a world in which abrupt changes seem to descend on us daily.
So why, in these ferociously challenging times, would computer science students vacate their comfort zones to knowingly embrace … another challenge?
From the perspective of the finalists in the 2024 North America Student Challenge (NASC), this “why” stems from three powerful benefits:
- Research experience
- The chance to meet and network with their peers and CS professionals
- All those unexpected insights that arise in the wilds beyond our well-traveled zones
Because NASC’s final round was also held in conjunction with the 2024 IEEE International Conference on Big Data in Washington, DC, these benefits were even more plentiful.
Hasan Mahmood, a CS graduate student and member of Purdue University’s three-person NASC-winning team, echoed many finalists’ views of these benefits, which included opportunities to “attend insightful talks, discuss emerging research with experts, and better understand how our work fits into the broader data science ecosystem.”
NASC 2024: A Starting Point
NASC was conceived by Saurabh Bagchi, a member of the IEEE Computer Society Board of Governors and a professor at Purdue University. Bagchi presented his proposal for the event to the IEEE CS board. After feedback and modifications, NASC was accepted and funded as a CS event; it was also co-sponsored by Adobe.
Bagchi said that his goal was “to energize the student community” by giving them real-world software and data analytics challenges to solve.
He also purposefully emphasized the crucial role of soft skills in tech today, further pushing contenders from their comfort zones by making the challenge’s final round a public presentation at the Big Data conference.
NASC’s Two-Part Structure
NASC started with a preliminary challenge. Stepping up to tackle it were 43 teams from institutions across the United States, from large public universities to smaller private universities and research-focused R2 institutions.
Participating teams could choose to tackle one or more of the three challenges, which academics and industry professionals formulated based on three criteria:
- Real-world relevance
- The availability of real datasets
- The ability of the solutions to be objectively judged
The three challenge problems were as follows:
- Predicting data center usage metrics, such as CPU or memory utilization
- Inferring latent user preference from conversations with a chatbot driven by a large language model
- Predicting the invocation rate of a serverless cloud application
Round 1 submissions were judged for accuracy and run-time/memory-use efficiency, as well as on the subjective factor of the solution’s elegance, including code clarity and conciseness.
Based on scores from round 1, the top three teams were invited to Washington, DC, where they presented their solutions at an open session at the IEEE Big Data Conference and fielded audience questions. Among those in the audience were the challenge judges:
These judges scored each team based on their presentation and their solution’s quantitative performance and novelty. The latter quality was something Bagchi said was perhaps the trickiest for teams to grasp.
“The criterion of novelty had to be interpreted carefully,” said Bagchi. “There was no expectation that the solutions would be novel from a research standpoint. Rather, submissions that adapted existing machine-learning techniques to the challenge at hand—or combined multiple algorithms in non-obvious manners—were rewarded.”
The 2024 NASC Winners
After combining their first and second-round scores, the final three teams were ranked as follows:
- Winner: Bilal Saleem, Hasan Mahmood, and Omar Basit from Purdue University
- First runner-up: Daniel Leeds, Harrison Huang, and Jonathan Mak from Rice University
- Second runner-up: Eliot Hall from San Jose State University (SJSU)
The NASC Experience: A View from the Finalists
Members from the three winning teams recently discussed the challenge they chose and how they approached it, as well as sharing their overall experiences at NASC and Big Data.
The Challenges and Their Approaches
Hall, the sole member of the second-runner up team, is a sophomore studying computer science at SJSU. He chose the problem of predicting data center usage and described his approach as follows.
“I initially tried classic statistical models before moving on to more powerful sequence neural networks—specifically, transformers—but still with limited results,” he said. “It was only when I realized that I could provide both the left and the right context … that the models started improving. After that, the biggest challenge was to ensure that my neural network didn’t overfit the data.”
The Rice University Team also chose the data-usage prediction problem. As Huang, a master’s student in computer science, noted, this problem is critical to ensuring reliable, efficient resource management, particularly in supercomputers.
The Rice team’s goal, he said, was to predict “missing resource usage data from a supercomputer—gaps that hardware errors could easily cause.” Huang said their biggest problem was navigating the vast literature on data imputation to identify the best machine learning models to implement for the challenge; to begin, they reviewed state-of-the-art methods.
“After rigorous testing and evaluation, we selected the two highest-performing models: PatchTST and LightGBM,” said Huang. “To further enhance the accuracy of LightGBM, we incorporated a long short-term memory (LSTM) network to ultimately introduce a novel data imputation network of our own: an LSTM–LightGBM hybrid network.”
The winning Purdue Team solved and presented on all three challenges. On the inferring late user preference from chatbot conversations challenge, Mahmood said that their solution emerged from combining various breakthroughs they experienced, including exploring how large language models reasoned and how the team could improve their communications with those models.
Participant Highlights from NASC 2024
For Hall, the SJSU sophomore, Big Data was his first conference. He said he learned a lot–including how to optimize his future conference experiences–and that meeting the other finalists was a huge highlight.
“I wasn’t expecting to even know who the other participants were until the awards ceremony,” Hall said, “so I really enjoyed having lunch together, getting to know their stories—and their approaches to the competition problems—and exploring the city together.”
Mak, a senior CS student at Rice, said the benefits of participating in the challenge started almost immediately:
“As someone with no prior research experience, participating in the NASC provided a great opportunity to see what academic computer science research was really like,” Mak said. “By collaborating with two masters students, I learned new machine learning techniques and I discovered a new side of CS I never had exposure to before.”
Mak also cited the conference—and the learning and networking it made possible—as a big highlight.
“I was able to catch myself up with the cutting edge of machine learning and big data as well as learn about working in academia—a career path I had never really considered before,” he said, adding that the opportunity to “explore our nation’s capital and the wonderful museums in the National Mall was a very underrated highlight of the entire trip.”
Like his teammate Mak, Rice CS graduate student Leeds mentioned learning about opportunities to do academic research as a career path to be a huge benefit of participating in NASC. Leeds also said that working on the challenge problem let him apply his machine-learning background to one of his target domains: high-performance computing.
“It was very interesting to work with different deep learning architectures such as the Transformer-based approach we used on a multivariate time series problem,” Leeds said, adding that the multivariate time series data’s relatively low dimensionality created unique challenges for working with high parameter size models. “Addressing these challenges during the training process highlighted the importance hyperparameter tuning and optimization techniques have when training complex models like a Transformer.”
Mahmood said the fact that the challenge tackled practical problems using real-world data gave him valuable insights into “how helpful and feasible advanced algorithms are under production-level conditions—a perspective rarely captured in a classroom setting.”
He also said that NASC helped reinforce the importance of thoughtful innovation presented in a clear, coherent way. “Unlike many competitions, it was refreshing to see NASC valuing holistic problem-solving and clear communication instead of just raw technical prowess.”
2025: NASC Goes Global
Bagchi said that the event was valuable for participants, students, organizers, and judges alike for numerous reasons. “What will stay with me,” he said, “was the personal interaction with the students and judges at the conference. I understood at some level the journey the students went through to participate and win in the event and their longer horizon goals.”
Next year’s challenge will maintain the successful two-part structure of the 2024 NASC, but Bagchi said it is also expanding its reach and adopting a fitting new name: the Global Student Challenge (GSC).
Learn More
Details about GSC will be unveiled in late Spring and will be announced in Tech News.
Like the RazorHack Challenge, NASC went from a concept to a life-changing event through the vision of Bagchi and his fellow board members Joaquim Jorge and Deborah Silver. If you have an idea for an event, submit a proposal to the CS Emerging Tech Grant fund.
