Nov. 29, 2023
White House Executive Order: Making AI Work for the American People
On October 30, the White House announced a new Executive Order (EO) that “establishes new standards for AI [artificial intelligence] safety and security, protects Americans’ privacy, advances equity and civil rights, stands up for consumers and workers, promotes innovation and competition, advances American leadership around the world, and more.”
The Department of Energy (DOE), with input from national labs, is in the process of responding to the EO on multiple fronts. Representatives from the DSI and AI Innovation Incubator (AI3) are working closely with LLNL leadership to provide the DOE with research results, successes, and strategies in AI safety. The Lab is home to many experts (see the story below on Bhavya Kailkhura as one example) whose projects tackle the security and reliability of AI and machine learning (ML) technologies—such as large language models, image classification algorithms, and model compression techniques—from all aspects of the AI/ML lifecycle.
In addition to the EO link above, the ai.gov website has been updated to reflect these national priorities. Of note is the workforce page, which lists LLNL’s Data Science Summer Institute as leading example of how students can enter this impactful space and build their skills accordingly. For more information about LLNL’s response to the EO, look for future editions of this newsletter and other LLNL communication channels.
Young Leaders Attend Science and Technology in Society Forum
Early-career staff scientists Kelli Humbird, Chris Young, and DSI director Brian Giera connected with Nobel Laureates and discussed important global issues ranging from AI to climate change at the 20th annual meeting of the Science and Technology in Society (STS) Forum in Kyoto, Japan. Lab Director Kim Budil, Acting Chief of Staff Ashley Bahney and Strategic Deterrence Associate Director Mark Herrmann also attended the annual meeting in early October, along with nearly 1,000 scientists and leaders from across the world in the fields of politics, business, and technology. “It is an honor for representatives of Lawrence Livermore National Laboratory, especially our young leaders, to participate in these important discussions about the impacts and implications of science and technology from a long-term perspective, looking far into the future,” Budil said.
“The meeting was a unique experience. Unlike many scientific conferences that are usually focused on a single field of research like plasma physics, this event had a diverse set of attendees. There were folks from all over the world, some researchers, some CEOs, government officials and more, so the conversations on specific topics were quite fascinating with such a wide array of perspectives,” said Humbird, a team lead for inertial confinement fusion cognitive simulations. All three of LLNL’s Young Leaders say they will carry forward the experience of the STS Forum into their daily work.
DOE Data Days Event Brings National Labs Together
Data researchers, developers, data managers, and program managers from the DOE national laboratories visited LLNL on October 24–26 to discuss the latest in data management, sharing, and accessibility at the 2023 DOE Data Days (D3) workshop. Sponsored by the National Nuclear Security Administration’s (NNSA) Office of Defense Nuclear Nonproliferation and hosted annually by LLNL, the event featured more than two dozen speakers from across the DOE/NNSA complex. More than 200 attendees met to explore wide-ranging topics in the data sphere, including common challenges the national labs face in operating in a secure environment and maintaining an ever-increasing amount of data generated in the age of AI, ML, and other advanced high-performance computing (HPC) technologies.
The workshop’s daily sessions covered data access, sharing and sensitivity, data curation and metadata standards, data governance and policy, cloud and hybrid data management, and data-intensive computing, with each session concluding with a question-and-answer period. Attendees shared their respective lab's latest data management tools and platforms, as well as best practices and possible avenues for improving data literacy and accessibility. Speakers from DOE, NNSA and the national labs also highlighted emerging trends being applied to DOE workloads across the complex, such as generative AI, large language models and converged computing.
Common threads throughout the workshop were the importance of building a “data community” across DOE, the need for a shared language around data, and the value of national lab personnel who are working on data issues. “Technology changes rapidly, which can both provide challenges and opportunities in our research; it is critical, as data managers across the spectrum of specializations, that we have a forum to address challenges, learn from one another, engage in thoughtful discussions, and drive change and innovation,” said Rebecca Rodd, event host and LLNL geophysics data specialist.
Bhavya Kailkhura Named IEEE Senior Member
The Institute of Electrical and Electronics Engineers (IEEE), the world’s largest technical professional organization, has elevated LLNL research staff member Bhavya Kailkhura to the grade of senior member within the organization. IEEE has more than 427,000 members in more than 190 countries, including engineers, scientists, and allied professionals in the electrical and computer sciences, engineering, and related disciplines. Just 10% of IEEE’s members hold the status of senior membership, which reflects professional maturity and requires extensive experience and documented achievements, according to IEEE. It is the highest grade for which IEEE members can apply.
A computer scientist in LLNL’s Center for Applied Scientific Computing, Kailkhura’s research interests include safe and trustworthy AI, efficient ML, and their application to scientific applications. “As an IEEE senior member, I am honored to be recognized for my contributions to the field of safe and trustworthy AI,” Kailkhura said. “This prestigious acknowledgment not only symbolizes personal growth but also underscores LLNL’s consistent commitment to pioneering research that ensures the development of dependable AI, aligning with [President Biden’s] recent AI Executive Order, for the nation’s security and scientific advancements.”
LLNL Partners with new Space Force Lab
LLNL subject matter experts have been selected by the U.S. Space Force to help stand up its newest Tools, Applications, and Processing (TAP) laboratory dedicated to advancing military space domain awareness (SDA). The Livermore team attended the October 26 kickoff in Colorado Springs of the SDA TAP lab’s Project Apollo technology accelerator, designed with an open framework to support and encourage rapid government, industry, academic, and allied collaboration. Project Apollo will focus on solving a variety of technical challenges, including developing capabilities to detect rocket launches using open-source seismic data; predicting the trajectory of rockets; and directing a sensor, such as radar, to search and track a rocket using predictive algorithms.
“Project Apollo was launched to inspire collaborative problem solving of critical space-domain challenges and facilitate the transfer of technology from private industry to Space Force operators,” said Jason Bernstein, principal investigator. “The Lab has a sweet spot for all fundamentals―data science, astrophysics, HPC, uncertainty quantification, modeling and simulation―that will enable Project Apollo’s success.”
Luc Peterson, associate program leader for data science in LLNL’s Space Program, added, “The big challenge of SDA is how to use all of the data efficiently to determine what’s orbiting and where it’s going. The data can be multimodal, such as optical images and radar signatures, from a variety of distributed sensors. Multimodal data fusion in an operationally relevant timeline is the goal. With the burst of commercial satellite data, and the advances in AI/ML models, there’s hope that these can be combined into useful information.”
(Photo at left: LLNL staff Jason Bernstein, Imene Goumiri, and Peterson with U.S. Space Force Major Sean Allen, the first chief of the SDA TAP Lab.)
For Better CT Images, New Deep Learning Tool Helps Fill in the Blanks
At a hospital, an airport, or even an assembly line, computed tomography (CT) allows us to investigate the otherwise inaccessible interiors of objects without laying a finger on them. Computer algorithms can piece together the many angle-dependent projections to resolve and reconstruct the object’s interior in three dimensions. Dozens of techniques exist to mathematically reconstruct an object in 3D, but unlike in theoretical research where high-fidelity data may be plentiful, real-world implementations of CT reconstruction encounter several obstacles that can produce image artifacts and degrade reconstruction results.
LLNL researchers Rushil Anirudh, Jayaraman Thiagarajan, Stewart He, Aditya Mohan, and Hyojin Kim collaborated with Washington University in St. Louis to create a state-of-the-art ML–based reconstruction tool for when high-quality CT data is in limited supply. As detailed in a recent paper—accepted to the 2023 International Conference on Computer Vision (ICCV), one of the most prestigious global computer vision events—the team devised a deep learning–based framework for predicting the absent CT data and enhancing 3D reconstruction.
The team’s Diffusion Probabilistic Limited-Angle CT Reconstruction (DOLCE) model was exhaustively trained on hundreds of thousands of medical and airport security x-rays to learn how to incrementally refine these images and restore missing data through the deep learning process of diffusion. DOLCE can handle large data distributions to tackle CT reconstruction tasks outside the initial medical and security imagery it trained on. It also can quantify the uncertainty associated with a particular reconstruction. By reporting the level of confidence that the reconstruction is correct, DOLCE is extensible to an array of imaging challenges of research and security concerns.
Seminar Explores Higher-Order Multi-Variate Statistics
The DSI’s November 8 seminar was “Higher-Order Multi-Variate Statistics for Scientific Data Analysis” presented by Dr. Hemanth Kolla of Sandia National Labs. In this talk, Kolla presented a motivation for use of higher-order multi-variate statistics (e.g., joint moments, cumulants) for scientific data analyses, both observational and computational. He also presented two topics where his team had recent success with the use of cokurtosis: rare (anomalous) event detection, and dimensionality reduction for stiff dynamical systems.
Kolla is a Principal Member of Technical Staff in the Scalable Modelling & Analysis department at Sandia National Laboratories. His interests lie at the intersection of HPC and statistical learning. He is currently working on projects involving tensor decompositions for various analyses, efficient forward propagation of parametric uncertainty in computational mechanics, and algorithm-based fault tolerance for HPC.
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).
Meet an LLNL Data Scientist
As a member of the Analytics for Advanced Manufacturing group, Aldair Gongora supports the Advanced Manufacturing Lab (AML), the Center for Engineered Materials and Manufacturing (CEMM), and the DSI with projects in data science and ML. First, he’s accelerating the design of additive manufacturing applications by combining ML and automated experimentation to select and conduct experiments without human intervention (“self-driving” labs). His other projects are focused on scaling up technologies in climate and energy applications as well as developing and using a modular autonomous research system (MARS) for biological applications. For both, he designs intelligent, adaptive frameworks to connect various information sources—e.g., chemical and materials data, device-level performance, technoeconomic-analysis predictions—across scales. Born in Belize, Gongora won a scholarship to Rockhurst University (Kansas City) for his bachelor’s degree, following up at Boston University with a PhD in autonomous experimentation for mechanical design. Gongora’s journey has taken him far from home, and he’s grateful for the support along the way. “I feel fortunate to have had conversations with many great people who inspired and mentored me during my education and here at LLNL,” he says.