Alum and machine learning pioneer presents a new picture of data science, with signal processing a critical component
By Patrick L. Kennedy
Arguing that signal processing deserves to be considered part of the canon of data science, Alfred O. Hero, III (ENG’80) delivered the 2026 DeLisi Lecture in Boston University’s Metcalf Trustee Ballroom on March 19.
The Charles DeLisi Distinguished Scholar Award and Lecture recognizes researchers with extraordinary records of well-cited scholarship, senior leaders in industry, and inventors of transformative technologies. The event gives the recipient a forum to discuss their work before the BU academic community and the general public.
The audience for Hero’s lecture, “Statistical Signal Processing and the Foundations of Data Science,” also included former students of Hero’s who are now in the Boston area. “Students are the ones that allow us to move forward with ideas which usually are only partially formulated,” said Hero, “and they’re the ones who actually do the work.”
An expert in modeling across fields
Hero knows a thing or two about being a hard-working student. For three of the four years that he was an undergraduate in the BU College of Engineering, Hero also worked full-time—on the night shift—at Raytheon Data Systems. After receiving his bachelor’s degree, summa cum laude, from BU in electrical engineering, he earned a PhD from Princeton University, also in electrical engineering, and went on to a long career at the University of Michigan, Ann Arbor.
At Ann Arbor, Hero is the John H. Holland Distinguished University Professor of Electrical Engineering and Computer Science and the R. Jamison and Betty Williams Professor of Engineering, with courtesy affiliations in the biomedical engineering and statistics departments. He also founded and co-directed the Michigan Institute for Data Science. And during a recent leave from Michigan, he served as a program director in the CISE Directorate at the NSF.
Hero’s areas of expertise include modeling high dimensional spatio-temporal data, multimodal data integration, statistical signal processing, and machine learning, in particular focusing on predictive mathematical models for the biological and physical sciences, social networks, network security and forensics, and personalized health and disease. He has helped develop algorithms with the potential to predict the behavior of microbes, the immune system, and even solar flares.
The signal processing view
In his talk, Hero first laid out the commonly held conception of data science, a sort of nesting-doll view (see figure at right) in which the data science field encompasses artificial intelligence (AI), which in turn includes machine learning (ML) as a subset, with neural networks (NN) being a further subset of ML. Hero then presented an alternative.
“I propose a different view of data science, which I call a signal processing view,” Hero said. Signal processing has been a critical component of neural network research since the early days of the field during World War II, Hero said, citing pioneers such as Bernie Widrow, Norbert Wiener, John Tukey, and Claude Shannon.
“I’d argue that these individuals should be as famous as Albert Einstein, because of the impact that they’ve had on technology, in particular on the early development of radar and localization,” Hero said. “Target localization, in particular, was really key to developing a rigorous mathematical representation of unstructured data.”
And signals are unstructured data, Hero explained. They require processes of amplification, digitization, reformatting, and feature extraction in order to interpret the data, especially by correlating similar signals. Signal processing is embedded in all manner of modern technologies, including those used in air traffic control, weather forecasting, speech recognition and synthesis, geophysical exploration (for mining, agriculture, archaeology, and more), cardiology, medical imaging, GPS navigation, and more.
Predicting space weather?
To take just one example from Hero’s myriad research projects, a Michigan team developed deep learning models to predict the intensity of solar flares. Periodic bursts of extreme auroral energy pose a frightening risk to our modern telecommunication systems. “A major solar flare of the X class, which is the highest intensity, can cause blackouts, where satellites cease working,” Hero said.
Hero and colleagues trained and evaluated two deep-learning algorithms on virtual stacks of archived images from the Solar Dynamics Observatory. The system focused on active solar regions and scrutinized images that were taken over the 12 hours leading up to a flare, Hero said, “to see if the pre-flare activity somehow contains a morsel of information” about the imminent burst of solar energy. “We can partition the flare event from the SDO data. So we can see where the flare occurred, how intense it was, and at what time it occurred.”
Using their algorithms, the team did “significantly better in terms of the average error at predicting a flare 12 hours ahead,” said Hero. Theoretically, when the same patterns appear in images being evaluated in real time, a system based on this method could detect a flare ahead of time, though such an early-warning system is “nowhere near” being available to officials and the public, Hero said.
Making models fair
More work remains to be done, too, in another area of Hero’s expertise, ensuring fairness and equity in AI’s use of health data. For example, “when an AI tries to cluster subjects or patients for a target drug trial, you don’t want the clusters to be unbalanced,” says Hero, who recently published a paper on the topic in the Electronic Journal of Statistics. “You don’t want to exclude all African Americans from the cluster; you want to make sure that the cluster proportions of different demographics match the population proportions. So by doing this, we can obtain a quantification of fairness in clustering.”
This is one of countless applications for signal processing, Hero emphasized. “From an engineering perspective, signal processing should play a role equal in the canon to other methods promulgated in AI that typically don’t have performance guarantees,” he said. “Signal processing methods can effectively transform the unstructured signal into ingestible data that an AI can work with. My point is there’s been inadequate recognition that unstructured data is a big problem that to a great extent can be solved, when the data comes from some natural source, through signal modeling, that allows you to incorporate that information.”
Hero is a Life Fellow of IEEE, past president of the IEEE Signal Processing Society (SPS) and past member of the IEEE Board of Directors. A Fellow of the Society for Industrial and Applied Mathematics (SIAM), he helped launch the SIAM Journal on Mathematics of Data Science. Hero has won several best paper awards, the 2013 IEEE SPS Technical Achievement Award, and the 2020 Fourier Award, among other honors.
Awards for early-career researchers
Also at the DeLisi event, Associate Dean for Research and Faculty Development Ayșe Coskun presented the Early Career Excellence in Research Awards to Assistant Professor of Mechanical Engineering Emma Lejeune and Assistant Professor Biomedical Engineering and Mechanical Engineering Brianne Connizzo. This award celebrates the significant, recent, high-impact research achievements of exemplary tenure-track faculty who are within 10 years of receiving their PhD.
Lejeune has established a nationally recognized research program at the forefront of computational mechanics, computational biomechanics, and open science. Her work addresses a critical challenge in modern engineering: ensuring that machine learning methods applied to mechanics are rigorous, generalizable, and reproducible. She has created foundational community resources—including benchmark datasets such as Mechanical MNIST and FEM-Bench, as well as widely used open-source software tools for image-based analysis of engineered tissues—that provide the infrastructure needed to evaluate and advance ML-enabled computational science. Her research bridges fundamental mechanics and modern data-driven approaches, enabling transparent, reproducible interpretation of complex experimental data across materials science and mechanobiology. Lejeune has secured significant funding from federal, industry, and foundation sources, and she has garnered major honors such as the American Heart Association Career Development Award and selection to the NAE Grainger Foundation Frontiers of Engineering Symposium. She is building not only powerful computational tools, but also the open scientific foundation that will shape the next generation of engineering research.
Connizzo has built a transformative research program in extracellular matrix biology, mechanobiology, and multiscale biomechanics. Her pioneering work develops novel, real-time tools to probe tissue structure and function, fundamentally advancing our understanding of how mechanical forces, aging, sex differences, and inflammation influence tissue remodeling and repair. By integrating proteomics, mechanics, and cell biology, she is establishing new paradigms in tendon biology and musculoskeletal health, with important implications for aging, injury, and women’s health. Since launching her independent lab, Connizzo has demonstrated exceptional productivity and impact, publishing extensively while securing major extramural funding, including a prestigious NIH R35 MIRA award. Her work is supported by federal agencies, foundations, and research alliances, reflecting both its scientific rigor and broad relevance. In addition to her research excellence, she is a deeply committed mentor and leader within her field, contributing to national review panels, editorial boards, and initiatives that advance women’s health research and leadership in engineering.
Father of the Human Genome Project
One of the college’s signature events, the DeLisi Lecture was endowed by Charles DeLisi, who is widely considered the father of the Human Genome Project and served as dean of the college from 1990 to 2000. DeLisi recruited leading researchers in biomedical, manufacturing, aerospace and mechanical engineering, photonics and other engineering fields, establishing a research infrastructure that ultimately propelled the college into the top ranks of engineering graduate programs. In 1999 he founded—and then chaired for more than a decade—BU’s Bioinformatics Program, the first such program in the nation.
During his career, DeLisi directed the Biomolecular Systems Laboratory, where more than 100 undergraduate students, graduate students, and post-doctoral fellows have trained. In recognition of his outstanding scholarship, he was named the Metcalf Professor of Science and Engineering at Boston University.