Getting Ready for Large-Scale AI Research with LUMI

When Shristy Shah started her industrial PhD at Aalborg University and Milestone Systems, she already knew that artificial intelligence and large video datasets would play a major role in her research. To prepare for this, she participated in the LUMI Summer School through DeiC and gained hands-on experience with high performance computing (HPC) and one of Europe's most powerful supercomputers, LUMI.

Preparing for AI research at scale

Shristy's research focuses on making AI-based video analysis more efficient. Rather than analysing every frame of a video in the same way, she is exploring how AI can use information already stored within compressed video files to identify which parts are worth analysing in more detail. The goal is to reduce the amount of computing power needed while still achieving reliable results. This could make it easier to analyse large amounts of video data and help researchers and companies work more efficiently with AI-based video analytics. As her PhD progresses, she expects to train increasingly advanced AI models on large video datasets. Running these types of experiments requires significantly more computing power than a standard workstation can provide.

First experiences with LUMI

Although not entirely new to HPC, Shristy had never worked with a supercomputer on the scale of LUMI before starting her PhD.

"The summer school was my first real contact with LUMI. My PhD had only just started and I already knew that LUMI would become part of my work as the project grew, so the timing was almost too good."

The LUMI Summer School is an annual training programme for PhD students who want to learn how to use large-scale computing resources in their research. For researchers working with AI, access to computing infrastructure such as LUMI can be transformative. Instead of waiting days or weeks for experiments to run, researchers can analyse larger datasets, test more ideas and accelerate their research.

"The real challenge is that research is never just one experiment. You test an idea, learn from the results and then try something new. Access to systems like LUMI makes it possible to do that much faster and explore more ideas within the same project."

For Shristy, learning directly from the people who run and support the system was one of the biggest advantages.

"Learning from people who work with the system every day seemed much better than trying to figure it all out on my own later. The topics matched almost exactly what I was about to start doing in my PhD."

Why HPC matters for research

As AI models become larger and datasets continue to grow, access to HPC is increasingly important for many research fields. Systems such as LUMI give researchers the computing power needed to tackle problems that would otherwise be too time-consuming or computationally demanding. For Shristy, this means being able to work with larger amounts of video data and iterate on her ideas more quickly.

"A single dataset can consist of hundreds of hours of video. With access to LUMI, I can test more ideas, compare more approaches and reach conclusions faster than I could on a local machine."

Learning and networking

Beyond the technical training, the summer school provided an opportunity to meet researchers from a wide range of disciplines and countries.

"That was one of the best parts. There were people from physics, chemistry, AI and several other fields. Their research problems looked completely different from mine, but we were all using the same infrastructure in different ways."

The combination of lectures, hands-on exercises and discussions with fellow researchers created a valuable learning environment and provided new perspectives on how HPC supports research across scientific fields.

A Recommendation for other PhD students

Shristy expects to use many of the skills from the summer school directly in her PhD project and encourages other researchers to explore HPC opportunities early in their research careers.

"If I could give one piece of advice, it would be to go early in your PhD rather than late. The step from a local workstation to a supercomputer is smaller than many people think, and the knowledge pays off immediately."

Every year, DeiC offers PhD students the opportunity to apply for the LUMI Summer School. Keep an eye on the DeiC website for upcoming opportunities. You can also follow researcher.aau.dk and the ReachAAUt Teams channel, where we regularly share relevant opportunities for AAU researchers and PhD students.

If you are interested in exploring whether LUMI could support your research, CLAAUDIA can help you understand the available opportunities and guide you through the application process. Contact CLAAUDIA here.