March 1, 2019
Machine learning might seem impersonal, but for Laura Mascio Kegelmeyer, the technology can strengthen personal relationships. “For me, it’s about the team experience,” she says. “I love getting specialists together to combine their strengths and brainstorm solutions.”
At the Women in Data Science (WiDS) 2019 conference, Kegelmeyer will share how her team has used machine learning to improve accuracy and efficiency for the optics recycle loop and how the technology has evolved since they first began using it in 2007.
Kegelmeyer founded the National Ignition Facility’s automated optics inspection analysis effort, which has been critical to efficient operations at the world’s largest laser. The overarching goal is to support the NIF Optics Recycle Loop in which optics damaged by NIF’s intense laser light can be removed from a beamline, repaired, and then re-used, thus extending their life.
Optics inspection involves looking for imperfections in NIF optics both on and off the NIF beamline. The idea is to quickly and efficiently distinguish sites of interest from irrelevant features using automated tools like image analysis and machine learning.
In 2016, Kegelmeyer conducted several machine learning efforts as part of an ambitious initiative to automate the optic repairs led by Mike Nostrand. That project eliminated the need for highly trained operators to painstakingly guide and monitor the repair of each damage site, which used to take eight hours per optic. They now spend only two hours per optic, freeing them to use their skills more broadly and efficiently. Automated inspection analysis now includes more than 22 production applications and continues growing and adapting to NIF’s dynamic needs.
To build these solutions, Kegelmeyer has invited collaboration from dozens of interns and seasoned experts alike. “Different machine learning methods have different strengths and new methods keep emerging,” she says. “From ensembles of decision trees to deep learning and even a combination of the two, we have carefully evaluated and adapted these to use the right tool for each challenge.”
Kegelmeyer considers her ability to collaborate one of her greatest strengths. “I’m not an expert on deep learning, but I don’t have to be. We have world-class experts right here at this Lab, so it’s more about building relationships and networking to pull together the right expertise to tackle the problem at hand,” she says.
Her husband is her longest-standing collaborator; they met and coordinated efforts towards breast cancer detection in 1993. How Kegelmeyer came to work on breast cancer detection is part of a longer journey that began in a remote village in central Italy, the home of her parents and two of her five older siblings.
After World War II the family was spurred to make the difficult move to the United States, where Kegelmeyer and the rest of her siblings were born and raised, with new futures available to them. Through hard work and perseverance, they earned college degrees in math, engineering, and medical fields.
Watching her older siblings pursue their education and careers, Kegelmeyer chose a path that combined all their specialties. She earned her bachelor’s degree from Boston University’s fledgling biomedical engineering program. The field was so new and unfamiliar to potential employers that she pursued a master’s in electrical engineering, also at BU. She was delighted to land a job at LLNL to study the physics of cells, mixing the “biomedical” back in with her engineering skills.
During her dozen years with LLNL’s biophysics program, Kegelmeyer researched topics such as genetic abnormalities and DNA probe mapping. She then created one of the first algorithms for computer-aided diagnosis of mammograms to differentiate between benign and malignant suspicious spots. Such developments have led to earlier breast cancer diagnosis and fewer unnecessary biopsies.
Among these significant technical achievements, Kegelmeyer also makes time to encourage the next generation of scientists. Nearly every summer she mentors summer students — as many as seven in one year. She enjoys working with students and is proud to note that a few former interns have become staff scientists at NIF. Her efforts to support ignition at NIF and to inspire others to pursue the same line of work are her legacy — a way to help future generations address national and global challenges.
Her advice to those students is to “play to your strengths.” Her definition of success is to “meet your goals and have fun doing it. I think we all have to chart our own paths and that starts with finding your passion.”
Outside of work, Kegelmeyer’s passions include hiking, reading, and dancing. She and her husband originally intended to take a few dance lessons just to prepare for their wedding dance, but they’ve been taking lessons, dancing and even giving lessons for more than two decades since.