5 Projects That Push the Limits of Physics, Fabrication Techniques, Algorithm Design
Two engineering professors among the NSF CAREER award recipients: William Boley and Francesco Orabona
By Jessica Colarossi for the Brink
![NSF 2021 Career Winners at BU](/eng/files/2021/07/Resize-NSF-636x425.jpg)
Five Boston University researchers have recently received Faculty Early Career Development Program (CAREER) awards from the National Science Foundation (NSF). The scientists are searching for the existence of dark matter, building smarter machine learning algorithms that don’t need human help, enabling better data analysis while protecting personal information, bringing multimaterial 3D printing techniques up to commercial scale, and making do-it-yourself sculptural design and fabric casting available to anyone with computer access. The Brink caught up with BU’s latest NSF CAREER award recipients to learn more about how their new funding will advance their work.
William Boley
During his PhD program at Purdue University, William Boley adapted an inkjet printer to 3D print electronics and biosensors. “I quickly learned how critical of a role the composition of the ink plays on the resulting geometries and properties of the printed structures,” says Boley, BU College of Engineering assistant professor of mechanical engineering and materials science and engineering. Inkjet printers are a common type of droplet-based fabrication—used to print text and graphics on paper—encountered every day in offices and homes around the world. This same type of approach is also used in various commercial 3D printing applications ranging from electronics to medicine.
One drawback of droplet-based printing, though, is that currently inks used in this process are only optimized for use on a single type of material. Scaling to printing multilayered structures made up of multiple materials and inks requires an additional printhead for each different ink, and additional time to sequentially deposit each layer, making multiple-printhead approaches prohibitively costly and time-consuming. With the support of his CAREER award, Boley is developing new 3D printing techniques for fabricating multimaterial objects and devices from a single printhead, which will make the process better positioned to be scaled up for commercial use.
“This technology will radically simplify the [printing] process,” Boley says. “Instead of integrating multiple printheads of single material inks…this technology will only require one printhead of a single ink of multiple materials.” This technology, he says, could be applied to any type of device that requires layers of multiple materials, leaving the door wide open for his technology to be adopted in a number of different fields.
Francesco Orabona
In machine learning and computer science, an algorithm is the set of rules that tells a computer or machine how to perform a task or function. But algorithms are reliant on humans to design them. Could machine learning algorithms become smart enough to not need humans to tune and refine them?
“Machine learning algorithms are not really automatic,” says Francesco Orabona, BU College of Engineering associate professor of electrical and computer engineering, systems engineering and computer science. “They need a lot of tuning and choices, typically made by humans.” In some cases, however, algorithm tuning could be done without human intervention. This is called a “parameter-free algorithm,” Orabona says, which is like having an oven determine its own optimal temperature and cook time to bake a cake.
Orabona started off developing vision systems for robots, but then decided to switch his focus to theoretical machine learning, drawing on his experience using algorithms in robotics. “That is when I started working on how to make these algorithms more and more automatic, trying to remove the human from the loop,” Orabona says.
With the CAREER award funding, Orabona hopes to design new truly automatic machine learning algorithms, making them “smart” enough to tune their own parameters without human intervention, saving people time and requiring less training time for the algorithm to function as intended.
Zeynep Demiragli
As the daughter of a physicist, Zeynep Demiragli grew up thinking and talking about science every day.
“I read everything I could get my hands on,” says Demiragli, a BU College of Arts & Sciences assistant professor of physics. “The more I read, the more I realized that I needed to learn about particle physics to really understand the nature of the universe and answer simple-sounding but quite complex questions: What is really out there? And of what is it all made?”
Demiragli’s research largely focuses on dark matter—a type of matter scientists theorize makes up most of the matter content in the universe—and works to discover and characterize how dark matter could interact with other types of particles.
“A promising idea is that dark matter could be made of a new type of particle, one that is not in the Standard Model,” she says, which is the set of rules physicists use to describe known particles and forces of nature like gravity and magnetism. With support from her CAREER award, Demiragli is continuing her search for the existence of dark matter here on Earth. She and fellow physicists are testing whether they can produce and observe dark matter in high-energy particle collisions at the Large Hadron Collider, the largest particle accelerator in the world.
“The collisions of tiny particles here on Earth could therefore tell us a lot about the origin and behavior of galaxies all across the universe,” Demiragli says.
Demiragli is bringing hands-on physics learning experiences to high schoolers, enabling students to build small particle detectors and learn the basics of accelerator physics. She also plans to organize workshops for undergraduate women in physics aimed at advancing their careers.
Emily Whiting
With a combination of computer science, imagination, and a DIY production process known as fabric formwork, computer scientist Emily Whiting and her team create large-scale, custom-shaped objects from fabric castings.
“3D printing isn’t practical for large-scale objects,” says Whiting, a BU College of Arts & Sciences associate professor of computer science. Fabric formwork—a technique advanced by Whiting that forms sculptural objects using fabric casts filled with setting materials like concrete—provides a cheaper, more accessible solution. Since it’s hard to predict the fabric tailoring needed to create a specific 3D form, Whiting uses computational methods to advance the process, a method that has since received attention from architects. With the CAREER funding, she is expanding the scope of fabric formwork by incorporating new techniques and computer graphics.
“We’ll explore new processes for fabrication that reduce the manual work in assembling the formwork,” Whiting says. Part of that will require her and her team to develop a system that can automatically generate visual guides for assembling the fabric, making the method more accessible to anyone interested in experimenting with forming sculptural pieces.
Whiting says the CAREER grant will also support graduate student research, new prototyping equipment, and an interdisciplinary workshop with architecture students. By the end of the five-year project, she hopes to have created a broadly accessible computational solution for constructing complex shapes and sizes using fabric formworks.
Mark Bun
The internet provides robust potential for gleaning insights from personal data that can transform medicine, social science, and technology, says Mark Bun, BU College of Arts & Sciences assistant professor of computer science. But without the proper tools to analyze data while safeguarding individual privacy, there remains untapped potential in the data.
“The overarching question I’m trying to address is, how can we learn population-level information from statistical datasets while safeguarding the privacy of the individuals who contribute data?” Bun says. With support from his CAREER award, Bun plans to continue developing the concept of “differential privacy,” which uses mathematical calculations to build privacy protection into algorithms that run on data.
“Once you’ve come up with a differentially private algorithm, it has the potential to solve problems in [analyzing] Census data, healthcare data, geolocation data,” Bun says.
However, for some especially sensitive data analyses, differentially private solutions might not be possible.
“I hope to establish a general and useful framework for deciding whether a statistical task can or cannot be performed with differential privacy,” Bun says. He also hopes to design and pilot a module for teaching these concepts in computer science at the middle and high school levels, so “a broad audience can come away with accurate intuition for how their data can be protected,” he says.