Institute for Computational and Data Sciences

ICDS engineers present, win best abstract award at national conference

Jamil Gafur, a doctoral student at the University of Iowa, and Gretta Kellogg, Penn State Institute for Computational and Data Sciences (ICDS) assistant director of strategic initiatives and director of the Center for Applications of Artificial Intelligence/Machine Learning to Industry, pose together for a photo at the PEARC conference. ICDS staff recently presented their research abstracts at PEARC, which is an annual high-performance computing conference bringing together industry partners, academicians, discipline experts and students together. Kellogg mentored Gafur through the conference. Credit: Gretta KelloggAll Rights Reserved.

UNIVERSITY PARK, Pa. — It’s not every day that researchers, academicians, industry partners, discipline experts and students in high-performance computing (HPC) can be a part of a forum to connect and discuss complex challenges, learn about software development and accessibility, grant funding and publication opportunities. 

The annual Practice and Experience in Advanced Research Computing (PEARC) conference allots attendees the space for these connections. Each year faculty and researchers from Penn State’s Institute for Computational and Data Sciences (ICDS) take part in presenting their work to this national audience and participating in conversations with our peer institutions and industry for the betterment of our shared research computing. This year, the conference took place in Rhode Island from July 21-25. 

Jeff Nucciarone, Research Innovation with Scientists and Engineers Team (RISE) engineer, won a Best Abstract-Student award for his work, “A Novel Approach in Using MPI to Manage a Coupled Workflow.”  

Message passing interface, or MPI, is the heart of the computation of your code, according to Nucciarone. MPI is used to move data around to perform calculations. His research started when he was working to help graduate students run models, who were running into problems with job scheduling. His research designed a way to automate the whole process using a less commonly used approach, called a parallel Fortran application, that uses a dynamic framework with multiple job runs to increase efficiency. 

“It felt nice to be recognized,” Nucciarone said. “What seemed like a temporary solution [to the student’s problem] blew up into a necessary research tool. Parallel Fortran applications are normally used as the heart and soul of numerically intense calculations, not as a job scheduler ... that was its novelty.” 

Lindsay Wells, system engineer, presented her research team’s abstract, “Leveraging Large Language Models for HPC User Support: A RAG-Based Chatbot,” to a full room of attendees and industry experts. 

The research team, comprising Wells, Amit Amritkar, assistant director for advanced computing; Simon Delattre, RISE engineer; Emery Etter, RISE engineer; and Justin Petucci, RISE engineer, worked on the research as part of the ICDS i-ASK help desk which answers questions regarding software installation, accessing resources, workflow troubleshooting and optimization.  

The retrieval-augmented generation (RAG) approach to the chatbot allows large language models as general learners. The researchers curate a local knowledge base of repeat questions, user guides and resource manuals, which allow the chatbot to answer the questions more accurately, improve access and usability, as well as reducing support burdens on client facing ICDS teams. 

“After my presentation, the number of people that approached me was amazing,” Wells said. “There was so much collaboration in the short time that I was there. ICDS having a general presence for networking with other people in similar roles as well as presenting allowed me to connect with other people who are trying to leverage large language models to enhance user support. Being able to present our talks, it connected us with people we would not have met otherwise.” 

Nucciarone and Wells attended the conference with Chad Bahrmann, assistant director of engineering and science services; Carrie Brown, ICDS advance cyberinfrastructure and education facilitator; Gretta Kellogg, assistant director of strategic initiatives and director of the Center for Applications of Artificial Intelligence/Machine Learning to Industry (AIMI). 

“It is my favorite conference,” Brown said. “It was the first conference geared towards academic research computing. The tools we use were developed by other units like ours that also attend these events. There are different levels of information you gather from attending ... professional and workforce development, policy, stakeholder and administration aspects. Given that this is the only platform to talk about research computing and the profession, it is a big deal to be recognized. It is just as prestigious as being published in a journal.” 

The week-long conference’s theme was “HPC: Human Powered Computing,” and featured a Globus tutorial for managing data flow, workshops and vendor meetings with industry experts, as well as Birds of a Feather (BOF) sessions, where groups of individuals with similar use cases come together to discuss the challenges that they face. 

“The Birds of a Feather sessions were useful,” Nucciarone said. “Sometimes it is the only time we get to meet with colleagues in person. A lot of ideas can take fruit from these interactions. There’s a lot of focus on AI and what AI can do. One of the interesting things I heard is that AI is new, and that you’re not as far behind as you think you are.” 

Kellogg also attended several workshops. 

“I spent an afternoon listening to talks by leaders from Indigenous Tribal nations," Kellogg said. "They talked about the research challenges they face as they develop the next workforce generation. PEARC gave their leadership a voice and an opportunity to present solutions to their own challenges.” 

Kellogg also mentors Jamil Gafur, a doctoral student at the University of Iowa. The two had originally met at PEARC in 2019. Gafur presented his work on a collaborative ICDS project with Cornell University using and developing various AI techniques to identify antibodies for various diseases. 

“His programming helps move the needle forward,” Kellogg said. “I love that not only is he a great student, but I am proud to help make some opportunities available to him. It has been great watching him grow. It is important for us to pass on our knowledge and these opportunities. It is incumbent on us to teach the new generation.” 

Nucciarone and Wells both highlighted the importance of the mentor program. 

“I’d like to see more student mentoring,” Nucciarone said. “Meeting the students was fantastic and seeing the work that they are doing. In Uganda, a professor who bult a small cluster at their university trained 20 students on the system. It was a whole new world for them.” 

PEARC is different from other workshops, according to Kellogg.

“It is very inclusive," Kellogg said. "The development of community is important. It represents Penn State well and reinforces and provides that visibility. It helps us make direct connections.” 

Penn State Institute for Computational and Data Sciences Research Computing Systems Engineer Lindsay Wells presents her team's research at PEARC in a lightening talk. Credit: Chad Bahrmann/ICDSAll Rights Reserved.

Last Updated September 10, 2024