Social Media Graphs Power SC25 Reproducibility Challenge

Reproducibility is at the heart of good science. This year, it’s also at the heart of the Student Cluster Competition (SCC) at SC25.

A paper presented at SC24 by researchers from the Georgia Institute of Technology has been selected as the Reproducibility Challenge problem for student teams. The work, “Asynchronous Distributed-Memory Parallel Algorithms for Influence Maximization”, explores how to optimize graph searches for social media. It was chosen for its scientific merit, compatibility across diverse hardware, and strong author engagement during the review process.

Bringing Science to Students’ Hands

At SC25, student teams will attempt to replicate a subset of the paper’s results on their own supercomputers. Because SCC systems are smaller than those used in the original study, the problem size has been scaled accordingly, but the scientific challenge remains.

As SCC Chair Le Mai Weakley explained:

Through the Reproducibility Challenge, students get a chance to understand the nuances and importance not only of reproducibility and replicability in science but how reproducibility and replicability in computing directly play a role in scientific discovery and the challenges around that.

This means that while students gain hands-on practice with cutting-edge code, they’re also confronting the same issues that professional researchers face when scaling work across different systems.

Why This Paper Matters

The Reproducibility Challenge Committee wanted a problem that students could connect with. After all, influence on social media is something they experience daily.

Co-chairs William Godoy and Tainã Coleman noted:

This year we wanted to focus on a problem that is accessible and relatable to students. The influence of social media is something they experience on a daily basis and connecting the challenge with a familiar topic will amplify the importance of high performance computing in our daily lives. Another crucial aspect is that networking and memory movement are the biggest bottleneck in computing and AI, and the next generation of experts will continue to address these limitations.

In today’s era of heterogeneous computing, the task of coordinating data and computation across different platforms makes the students’ work particularly relevant.

How the Paper Was Selected

Choosing the right paper was no small feat. The committee, made up of experts from universities, national labs, and organizations across six countries, followed a rigorous process:

1. Initial screening to determine overall suitability for the competition.
2. Detailed reviews, with at least three evaluations per paper, to assess which applications were best suited for SCC teams.
3. Shortlisting and interviews with the authors to confirm feasibility and secure their commitment.
4. Final evaluation of reproducibility reports, hardware feasibility for SCC systems, and accessibility of the science to undergraduate students.
   

Beyond technical merit, the committee weighed reproducibility reports of the original code, accessibility for undergraduates, and the ability to run on SCC hardware.

Beyond Replication: Building Reproducibility Artifacts

The competition isn’t just about rerunning code. Each student team will produce its own reproducibility artifacts — data, scripts, and documentation — that showcase how they validated the paper’s results on their chosen systems. These outputs become a shared resource for the broader community, advancing the practice of reproducibility in high performance computing.

As Godoy and Coleman put it:

Scientific reproducibility challenges are something we see in all of our discussions. Students will learn the scientific value of discovery through computing. Like in any other science, the teams will aim to reproduce the data that led to new knowledge and validate conclusions.

Looking Ahead