For almost 70 years, Milt Halem, now a research professor at UMBC after a career at NASA, has deployed the latest computing technology—from vacuum tubes in the ’50s to artificial intelligence today—to tackle some of science’s toughest problems.
Colorful screen prints cover the walls of Milt Halem’s Rockville, Maryland, apartment. One shows a high-altitude view of the Nile River, cutting through desert sands before emptying into a turquoise Mediterranean. In another, the fragile blue marble of the Earth rises above a desolate lunar surface.
One large picture, taking centerplace on the dining room wall, is more enigmatic—filled with alternating red, yellow, and blue patches. “This is what’s known as the quasi-biennial oscillation,” Halem explains.
Tens of thousands of feet up, in Earth’s stratosphere, the winds around the equator regularly shift direction, blowing east for many months, and then west, and then east again. A satellite measured the shifts, and Halem, a former NASA scientist and current UMBC professor, turned six years of that data into art.
Halem’s framed prints, many of which he created himself while taking evening and weekend classes at the Corcoran School of the Arts and Design in Washington, D.C., are a window into his nearly 70-year-long scientific career. For roughly 40 years of that time, he worked for NASA, the famed federal space agency. Halem harnessed NASA’s space-based instrument “eyes” and ever-evolving computer “brains” to advance humanity’s understanding of Earth and our place in the cosmos and, in 1996, was recognized with NASA’s highest award, the Distinguished Service Medal.
In 2003, after retiring from NASA, Halem joined UMBC as a research professor. In his more than two decades at the university, he has nurtured connections with government and industry, brought in research money and equipment, regularly taught a popular graduate course, and mentored more than a dozen Ph.D. students, helping them pursue their passions and propel their careers.
“I think of myself as a computational scientist: I use computers to both discover and explore new territory,” Halem says. “What I love is the computer’s ability to represent, and to explain, the world around us.”
Charting computers’ evolution
Halem began his working life at the dawn of the computing age, and his career trajectory mirrors the steady rise of those digital devices. After earning a bachelor’s degree in mathematics from the City College of New York in 1951, Halem was drafted into the Korean War effort and served on U.S. Army bases for two years, first in New Jersey and then in Arizona. After a stint clearing rocks off the golf course so that the officers could play, Halem’s skills were better put to use on some of the Army’s first attempts to use what at the time were relatively new-fangled devices called computers to improve surveillance of battlefields.
In 1955, Halem received a fellowship to work with an IBM 704 computer installed at New York University. The computer took up significant floor space with its multiple cabinet-sized components, filled with vacuum tubes for logic operations and magnetic tape for memory. It was state-of-the-art in its day, able to execute about 40,000 instructions per second.
“You could say 1955 was my start in high-performance computing,” Halem says. “I’ve been acquiring and managing and programming and doing research on computers ever since.”
In the late 1950s, NASA established the Goddard Space Flight Center (GSFC) and, shortly after, opened a New York campus—the Goddard Institute for Space Studies—focused on theoretical research.
Milt Halem delivers remarks during a conference at NASA Goddard Space Flight Center, 1986. Photo by NASA.
Because of Halem’s experience with computers, he was hired as contract support. In the subsequent years, he worked with some of the most advanced computers of the time, using them to model the Earth’s atmosphere and predict the scientific value of launching new Earth-observing instruments. In 1968, he also earned his Ph.D. in applied mathematics from NYU.
Halem at UMBC in 2024.
In the late ’70s, Halem moved to Goddard’s main facility, GSFC,
in Maryland to continue his computing work. In 1983, he became chief of the Space Data and Computing Division and shifted from using supercomputers to being a force in their development. The division had recently developed the Goodyear Massively Parallel Processor (MPP), a trailblazing computer containing 16,384 processors working simultaneously. This degree of parallelism, which lies at the heart of today’s supercomputers, marked a radical departure from traditional computing.
“The MPP was a relatively simple concept, but it took us more than a decade to develop it,” says Jim Fischer, who managed the MPP’s fabrication contract and applications programs. “Milt found people to help and money to support the machine. And then he offered it to the U.S. science community to try it out, and they recognized its architecture as a solution. Milt pumped us to world class by force of will.”
In the ensuing years, Halem and Fischer pursued making the power of MPPs more accessible and less expensive. In 1994, their staff demonstrated the “Beowulf” computing cluster, composed entirely of mass market personal computers networked with Ethernet and running the Linux open source operating system.
Now 30 years later, the Beowulf approach sits at the core of most of the world’s supercomputers and much of the cloud infrastructure, putting vast amounts of affordable computing resources at the fingertips of scientists, engineers, and even social media users.
After Halem arrived at UMBC in his 70s, he continued to engage with the latest computing technology. He was instrumental in securing the donation of high-performance IBM computing equipment from NASA to the university, and he helped run a new computing research center. With his students, he has explored the frontiers of quantum computing and AI.
“I’ve followed the evolution of high-performance computing my entire career,” Halem says.
Computing for science
Computers can do many things—from recommending videos you might like to rendering life-like graphics—but Halem is most interested in what they can do for science.
“Science is based on the notion of observations and then explaining the observations by some theoretical process,” says Halem. “That was the way science evolved until the mid-20th century. Then computers came along and became the third pillar of science.”
Halem’s passion fit with NASA’s mission. Though the agency was born out of the Cold War and the space race with the Soviets, it has concentrated its efforts on the peaceful use of space to advance science.
NASA launches satellites that observe Earth from space with a suite of scientific instruments (including some designed at UMBC), and a large part of Halem’s research has been using computers to understand how that data can improve our understanding of the Earth’s air, water, and land and the physical processes that link them. Halem has also advanced weather forecasting through, among other techniques, finding new ways to incorporate observations into the weather models.
As an administrator at NASA, Halem always put the science first, according to his colleagues. “I believe that Milt left a legacy of making decisions that are in the best interest of science,” says Dan Duffy, the chief of the Computational and Information Sciences and Technology Office at GSFC, who sits in Halem’s old NASA office and has continued to work with him on special projects since Halem joined UMBC.
“His motivation was always driven by the scientific community’s needs,” Fischer adds. “It was very high-minded.”
Shaping students’ lives
At UMBC, Halem has continued his devotion to computing for science’s sake. In doing so, he has supported the research community and shaped his students’ lives.
Jennifer Sleeman, Ph.D. ’17, computer science, was co-mentored by Halem on her Ph.D. thesis, which involved mapping topics from past climate reports from the Intergovernmental Panel on Climate Change and trying to predict the contents of future reports.
Jim Fischer, Milt Halem, and Sophia Hamer (left to right) in Halem’s UMBC office. One of Halem’s screen prints hangs in the upper right.
“Milt and I would have discussions about my thesis, and I always like to describe it as: Milt decided to adopt me. I was close to the end of my Ph.D., and I changed my focus based on working with him,” she said. “It really changed the trajectory of my career.” After graduating, she became a research assistant professor at UMBC. She currently studies weather and climate-related topics as a senior AI research scientist at Johns Hopkins Applied Physics Laboratory while she continues to work part-time with Halem and his students.
Sophia Hamer ’22, mathematics and computer science, who is currently working with Halem as a master’s student, also credits Halem with helping her find her career footing. She first met him to discuss a research project when she was an undergraduate and just emerging from a time when she struggled with her grades and with choosing a major.
“Dr. Halem took a chance on me,” she says. “I honestly think that was the turning point for me. Before I was thinking, ‘I’m just gonna squeak out of here with a degree and hopefully my GPA isn’t too bad to get a job.’ But now, my entire perspective has shifted: I’m about to graduate with a master’s, and I’m doing impactful research in a field I really like. I never thought that would be possible.”
“Milt is tough—he pushes you—but he also works closely with you,” Sleeman adds. “I really grew from that relationship.”
AI for weather and climate forecasting
Both Sleeman and Hamer are working with Halem on a project to be the first to use AI to make regional weather predictions, one of a couple AI projects Halem is currently pursuing.
In his embrace of AI, Halem is yet again advancing alongside the technology that has defined his career. Gone are the days of mere 40,000 instructions per second. Now advanced computer chips power computationally hungry “neural networks” originally inspired by the workings of the human brain. Neural networks are behind celebrity chatbots such as ChatGPT—and forecasters are increasingly turning to them to predict the weather too.
While traditional weather models simulate the physics of the Earth and its atmosphere to make predictions about the future, AI forecasting models simply examine vast datasets of past weather and learn how to spot patterns.
So far, the main advantage of AI forecasting is speed. What can take days for a physics-based model can be done in only minutes or seconds with AI. AI weather models offer great promise for distributed, real-time weather forecasting, for example as natural disasters unfold. Halem and his students are looking in particular to predict wildfires.
While much AI weather forecasting research is undertaken by leading tech companies such as Google and Microsoft, Halem thinks his students are more than holding their own.
“I’ve got a new generation of computer scientists making breakthroughs,” says Halem. “Two students who are writing up their master’s theses are doing leading AI computations to rival what some big tech companies are doing.”
The work of weather and climate modeling is growing in importance as global warming unfolds and the Earth’s natural patterns shift. “Applying what I know to a real-world problem has been an amazing experience,” Hamer says.
The projects are also yet one more demonstration of Halem’s constant intellectual evolution.
“Milt has an amazing ability to learn new things,” says Anupam Joshi, the acting dean of the College of Engineering and Information Technology. “He was trained as a mathematician who then specialized in high-performance computing and in the last few years has become an expert in using AI systems to model weather and climate phenomena. It speaks to his intellectual capabilities but also his perseverance and his grit.”
Halem (far right) poses with (from left to right) NASA’s Jim Fischer and John Dorband, and Microsoft Fellow Jim Gray next to the first Beowulf computing cluster. Photo by NASA.
Into the future
Now in his tenth decade of life, Halem shows few signs of slowing down. When not spending time with his large family, he’s still tackling some of science’s thorniest problems, such as how to predict weather patterns months to years in advance.
“He’s like the energizer bunny,” says Fischer, his former NASA colleague.
Duffy agrees. “He has an amazing amount of energy; he never seems to stop. He has a willingness to listen, learn, and ask questions and is an amazing role model for his students.”
Halem says it’s the computation challenges that drive him. “I have to keep moving because the computational instrument continues to evolve,” he says. He continues to be on the lookout for new ideas—the latest he’s come across being an AI research scientist. “The AI agent can use large language models to read the literature; it can write computer code and conduct experiments by itself. It can write the paper on what it discovered. So that’s probably the ultimate research goal that I would like to see if I can make an impact on,” Halem explains.
“Milt is probably one of the most dedicated people I know to the research,” Sleeman says. “He’s so tenacious, and it’s just so impressive. It’s a privilege to continue to work with him.”
“Milt is incredible,” adds Joshi. “If we could replicate him, we would.” And maybe, in some sense, Halem’s AI scientist would do just that.
Tags: AI, cfr, COEIT, Fall 2024, Feature, GSFC, Mathematics, NASA
