Volume 30

In September, the DSI leadership team gathered for a two-day retreat to discuss goals, strategies, and activities for the next several years. The team consists of the director, deputy director (see next story), the council, directors of both student programs, data science ambassadors, and administrative and communication leads. As the hub of the Lab’s data science community, the DSI must evolve as LLNL’s mission space and the field evolve.

Some of the DSI’s plans for the new fiscal year, which began on October 1, include refining the Data Science Summer Institute and Data Science Challenge student programs, inviting more LLNL staff to become involved in the DSI, and documenting the next-five-years vision. One key initiative is expanding learning opportunities for staff through a machine learning (ML) training program co-sponsored by LLNL’s Center for Applied Scientific Computing. Longer term, DSI leaders are interested in avenues for data sharing among the national labs, procurement of additional high-performance computing (HPC) resources, and international outreach activities.

The retreat also aligned the DSI’s vision with that of the AI Innovation Incubator (AI3), which helps LLNL’s research programs engage with commercial and academic partners. “The DSI and AI3 are complementary in a number of ways. The former develops expertise and community, while the latter connects the Lab to public and private partnerships,” explained DSI director Brian Giera. AI3 director Brian Spears pointed out, “AI and related technologies are transformational for national security. The Lab brings a unique perspective and unique tools to bear on the application landscape.”

The retreat gave the team a chance to thank Ana Kupresanin (second from left in this photo), a member of the DSI Council since its inception in 2018. This month she began serving as director of Lawrence Berkeley National Lab’s Scientific Data Division. Giera stated, “Ana provided tremendous guidance during her tenure with the DSI. She is highly knowledgeable and respected across LLNL and the greater data science community, and we look forward to strengthening our collaboration with Berkeley Lab thanks to Ana’s new role.” In addition to DSI Council member, Kupresanin was deputy associate director for LLNL’s Weapon Simulation and Computing Computational Physics program.

LLNL data scientist Cindy Gonzales has been selected as the DSI’s first deputy director. This is a new role within the DSI leadership team, focused on working collaboratively to develop, implement, and promote LLNL’s strategic vision in data science, both internally and externally, as the DSI continues to grow.

Gonzales has been immersed in the data science community—both at the Lab and beyond—since 2016, when she became inspired by an ML seminar. She began working at LLNL that year as an administrator with the Computing Scholar Program, and has since built a data science career by interning with researchers and mentors through the DSI’s Immersion Program as well as earning a B.S. in statistics from Cal State East Bay and an M.S. in data science from Johns Hopkins University. Gonzales completed both of these degrees while working full-time at the Lab. In addition, she is committed to fostering diversity and inclusion institutionally and more broadly in STEM, earning a Diversity and Inclusion certificate from Cornell University in 2022.

Today Gonzales leads the Intelligent Detection, Exploitation, and Analysis (iDEA) Group in the Global Security Computing Applications Division, where her research interests include using ML for automatic target recognition and detection in unconventional types of imagery (e.g., overhead, radar, medical). In addition to her technical work, she has co-directed the Data Science Challenge, an intensive two-week student program. Read more about her career journey on the LLNL Computing website.

The Data Science Summer Institute (DSSI) hosted summer student interns from Japan on-site for the first time, where the students worked on real-world projects in AI-assisted bio-surveillance and automated 3D printing. From June to September, the three students—Raiki Yoshimura, Shinnosuke Sawano, and Taisei Saida—lived in rental apartments near the Lab and worked on different data science projects using electronic health records and neural networks trained on experimental data.

Sponsored by Japan’s Agency for Medical Research and Development (AMED) and the Japan Science and Technology Agency (JST), the relationship between the DSI and the Ministry of Japan stems from a series of agreements that arose in the wake of the Fukushima nuclear accident, to conduct academic exchanges and expand scientific collaboration. The DSSI opportunity began in 2019 but had to go all-virtual due to the COVID-19 pandemic.

The students were mentored by LLNL researchers Priyadip Ray, Aldair Ernesto Gongora, Andre Goncalves, and Jose Cadena-Pico. The mentors said they found the experience just as valuable as the students did. Gongora, who came to the Lab as a foreign national himself, said he resonated with the cultural challenges the students faced and said the opportunity epitomized the strength of diversity in science. “The benefit for me has really been being able work with someone from another country and learning more about Japan; learning about their lives, how their academic journey differs from education here in the U.S., and really finding the commonalities and differences,” he said.

The DSSI is accepting applications for summer 2024 until January 24. Interested students can apply online.

LLNL’s Physical and Life Sciences (PLS) organization is home to experts in materials science, physics, nuclear and chemical sciences, biosciences and biotechnology, and research on the Earth, its atmosphere, and energy—as well as the massive datasets generated by experiments and simulations in these fields. PLS recently convened a workshop exploring avenues for continued collaboration and innovation in leveraging AI tools, not only for making sense of multimodal data but also for accelerating scientific processes.

Sonia Létant, PLS principal deputy director, noted in the event’s opening remarks, “A growing number of the Laboratory’s research projects use AI to analyze unprecedented volumes of data and generate insights that enable faster and better research results.” Accordingly, AI3 director Brian Spears was on hand to discuss the unique challenges of developing and implementing AI in the Lab’s research environment. He stated, “We can’t use existing open AI tools to address LLNL’s high-consequence mission needs. Instead, we’re developing AI tools specifically designed to address our mission-driven research at the scale we need.”

The workshop highlighted opportunities to expand data science– and AI-related efforts already under way. Customized AI tools that leverage LLNL’s supercomputing resources assist PLS research on many fronts, including fusion energy, astrophysics, material optimization, energetic materials, and climate modeling. In addition, PLS researchers have mentored student interns through the DSSI and released public datasets through the DSI’s Open Data Initiative.

Since 2018, the DSI seminar series has showcased a wide range of work by nearly 60 external speakers—from academia, industry, and other labs—and LLNL researchers. The series is organized by a member of the Lab’s technical staff, supported by administrative and communications staff. As the new fiscal year begins, Leno da Silva will serve as the DSI’s next seminar series coordinator.

Leno joined the Lab in 2021 as a reinforcement learning researcher. He is a guest editor for Neural Computing and Applications and has organized several workshop series, including the Adaptive and Learning Agents workshop during 2020–2022 and the Latinx in AI workshop at NeurIPS. He has published one book and over 50 scientific papers in peer-reviewed venues. With a Ph.D. in Computer Engineering from the University of São Paulo, Brazil, Leno’s research interests include transfer learning, multiagent systems, power electronics, smart grid, transportation, antibody development, and AI-assisted healthcare.

“I am honored to have been selected for this role, picking up the torch after great work done by Sarah Mackay. I am excited to bring my own perspective to the table and help promote new conversations with academic and industry partners,” says Leno. DSI director Brian Giera adds, “Our seminar series and forthcoming conferences will benefit greatly from Leno’s perspective, reputation within the data science community, and approach to introducing fresh content to LLNL.”

See the next story for information about past and future seminars, including how to attend.

At the DSI’s September 7 seminar, Dr. Doug Bowman from Virginia Tech presented “Quantifying the Benefits of Immersion in Virtual Reality.” He reviewed decades of research on the benefits of immersion in VR, discussed hypothesized benefits alongside examples of empirical studies that provide quantitative evidence for these hypotheses, and talked about how successful VR applications can be applied in areas such as scientific visual data analysis.

Bowman is the Frank J. Maher Professor of Computer Science and Director of the Center for Human-Computer Interaction at Virginia Tech. He is the principal investigator of the 3D Interaction Group, focusing on the topics of 3D user interfaces, VR/AR user experience, and the benefits of immersion in virtual environments. Bowman is a co-author of 3D User Interfaces: Theory and Practice. He received his M.S. and Ph.D. in Computer Science from Georgia Tech.

Speakers’ biographies and abstracts are available on the seminar series web page, and many recordings are posted to the YouTube playlist. To become or recommend a speaker for a future seminar, or to request a WebEx link for an upcoming seminar if you’re outside LLNL, contact DSI-Seminars [at] llnl.gov (DSI-Seminars[at]llnl[dot]gov).

Although the electrocardiogram is a highly useful and commonly used diagnostic tool, there are limits to the types of information it can reveal. Pushing beyond those limits was the real-world, data-science problem future engineers from the University of California (UC), Riverside’s Marlan and Rosemary Bourns College of Engineering were tasked with exploring this summer at one of the foremost research facilities in the world. A group of 35 students from UC Riverside and UC Merced took part in LLNL’s annual Data Science Challenge. The intensive, two-week internship places students into multidisciplinary teams, each with a data scientist and a Ph.D. student as a team lead.

“This experience is targeted towards students who have little to no prior experience in data science and machine learning, and is meant to serve as a ‘crash-course’ in introducing them not only to the basic concepts, but also what it is like to be in the shoes of an LLNL scientist,” said Vagelis Papalexakis, an associate professor in UC Riverside’s Department of Computer Science and Engineering and internship facilitator. “I have seen this to be a very effective way of introducing students to data science, and by the end of the program it is impressive to see the student growth, especially in those students who started out with little to no experience.”

Grace Gu, a faculty member in mechanical engineering at UC Berkeley, has been selected as the inaugural recipient of the LLNL Early Career UC Faculty Initiative. The initiative is a joint endeavor between LLNL’s Strategic Deterrence Principal Directorate and UC national laboratories at the University of California Office of the President, seeking to foster long-term academic partnerships and provide UC faculty members with funding and Lab support for their research. The winning recipient receives up to $1 million in funding over five years to support an innovative research project in AI/ML.

Gu’s winning research proposal, titled “Toward AI-driven additive manufacturing for metal-ceramic composite structures,” seeks to develop new composite materials with exceptional properties, particularly in ultra-high-temperature ceramics for energy and defense applications. The project will focus on advancing the capabilities of binder jet 3D printing and optimizing composite feedstock development. This work is strategically aligned with applied data science efforts in materials and advanced manufacturing at LLNL.

Reduced order models (ROMs) combine data and underlying first principles to accelerate physical simulations, reducing computational complexity—and therefore computational costs—without losing accuracy. Preprints, full text, PDFs, or abstracts are linked where available.

• A fast and accurate domain-decomposition nonlinear manifold reduced order model – Youngsoo Choi with collaborators from Rice University (Image at left: Sparse autoencoder in which the input layer and decoder output layer are sparsely connected, and only the blue-outlined hidden nodes are required to compute the hyper-reduction nodes.)
• Data-scarce surrogate modeling of shock-induced pore collapse process – Siu Wun Cheung, Choi, and H. Keo Springer with a Sandia National Labs collaborator
• Epistemic uncertainty-aware Barlow twins reduced order modeling for nonlinear contact problems – Choi with collaborators from Sandia National Labs and Cornell University
• gLaSDI: Parametric physics-informed greedy latent space dynamics identification – Choi and Jonathan Belof with collaborators from UC San Diego and University of Arizona
• GPLaSDI: Gaussian process-based interpretable latent space dynamics identification through deep autoencoder – Choi, Debojyoti Ghosh, and Belof with a Cornell University collaborator

An annual tradition in the HPC community is the Supercomputing Conference—this year nicknamed SC23—to which LLNL regularly dispatches more than a hundred researchers, developers, and other HPC experts. Again this year, the largest supercomputing conference in the world will include data science–related sessions organized by or featuring LLNL staff:

• November 13: AI Assisted Software Development for HPC (AI4DEV) – Giorgis Georgakoudis, Ignacio Laguna, Konstantinos Parasyris (organizers). This workshop will explore how AI can assist with the HPC software development process.
• November 13: In-Situ Analysis and Visualization with Ascent and ParaView Catalyst – Cyrus Harrison and Nicole Marsaglia (presenters). This tutorial will introduce the in-situ visualization paradigm and walk participants through two open-source in-situ processing tools.
• November 15: Toward a National Artificial Intelligence Research Resource for Strengthening and Democratizing AI R&D – Fred Streitz (co-leader). This birds-of-a-feather session will provide a forum for strengthening the AI research and innovation ecosystem.

If you’re in Denver for the conference, stop by the Department of Energy’s booth to see a display about the role of cognitive simulation in LLNL’s fusion ignition milestone. LLNL’s full calendar of events is available online.

Data scientist Giselle Fernández contributes to an “exhilarating and immensely rewarding” breadth of LLNL research. She serves as ML lead for projects involving fusion energy design optimization, material deformation, and post-detonation flow transport. “Machines undoubtedly will play a pivotal role in our future,” she says. “My deep involvement in data science makes me feel connected to that future in an unprecedented way.” After completing an aerospace engineering Ph.D. at the University of Florida and postdoctoral research at Los Alamos National Lab, Fernández came to Livermore in 2020, joining the Atmospheric, Earth, and Energy Division to apply expertise in ML techniques and uncertainty quantification to research utilizing HPC resources. “Being part of the LLNL community affords me the honor of working for the safety and security of our nation—a responsibility I take immense pride in.” Dedicated to outreach and mentorship, Fernández regularly presents workshops and tutorials, authors news articles, and leads interdisciplinary ML discussions. She also hosts students every summer at Livermore. “My mom, who devoted her career to teaching early on, always fascinated me with her passion for education. Now, hearing students say, ‘I learned so much from you,’ I appreciate how deeply rewarding it is.”