One afternoon, Alan Willsky realized something that would come to be an essential part of how he approached research.

The young MIT faculty member was a consultant to a terrific team of engineers at The Charles Stark Draper Laboratory that was charged with developing the algorithm set for sensor fault detection for NASA’s Digital Fly-By-Wire (DFBW) aircraft. The goal was to make sure that sensor faults would be detected before faulty signals corrupted the fly-by-wire flight controls of the aircraft. This involved comparing data from the jet's multiple sensors in order to tell if one or more of them were giving an erroneous reading. While in principle this could be done using methods Alan and others had developed for dynamical systems and failure detection, different parts of the model had different sources of error, which complicated things. What the team did instead was to “rip the dynamics apart to isolate where they could get the most robust sources of analytical redundancy and then exploit these using the machinery of failure detection.”

The approach worked “phenomenally well,” resulting in reliable and accurate fault detection. But “it was so systematic, that I felt there must be a theory underlying this,” Alan says.

That realization, and the search for the accompanying theory, opened up a new avenue of research and led to dissertations and a master's thesis for several of his students. It was also a reminder to the young scientist that science must serve the real world. “The algorithms being developed at Draper Lab were going to be flight-tested,” says Alan, “so they were going to have to work. We had to be our own devil's advocate.” (As it turns out, the DFBW project ran until 1985 and resulted in the all-electric flight control systems now used on nearly all modern aircraft. In their project summary on the NASA web site, they say the program is considered “one of the most significant and most successful NASA aeronautical programs since the inception of the Agency.”)

Alan has spent close to five decades at MIT discovering how the practical informs the theoretical, and vice versa. Warm and loquacious, he will claim that he was merely in the right place at the right time. In fact, colleagues and students will say he has an extraordinary ability to see the connections between things, and it's this strength that has helped him thrive.

Early on, Alan says, he had the right sort of encouragement and good luck. The math-loving boy from New Jersey arrived at MIT in 1965, majoring in Aeronautics and Astronautics, otherwise known as 'AeroAstro'. (The fan of the New York Giants baseball team – a rival to the Yankees – was a natural Red Sox fan on arriving in Cambridge.) With the US space program booming, it was a thrilling time. The AeroAstro students watched the live launch of Apollo 10 in 1969, for instance, sitting in the viewing area with the King and Queen of Belgium and then-Vice President Spiro Agnew.

He chose the department for its diversity and flexibility. “If you think about it, AeroAstro covers all the branches of engineering; I got the opportunity to see a wide variety of different topics.” In his junior year, he took classes in probability, signals and systems, and control, which turned out to be his passion and forte. Advisors like Harvard's Roger Brockett were instrumental in helping him pinpoint which questions to answer, breaking down a big problem into simpler cases and coming up with questions which were solvable but which also helped advance the state of knowledge.

After a PhD in control and estimation, supervised by Brockett and Professor Wallace Vander Velde, Alan eventually arrived at MIT’s Department of Electrical Engineering and Computer Science in 1973. “At the time, LIDS was way over on the corner of Vassar and Mass Ave, while CSAIL [the MIT Computer Science and Artificial Intelligence Laboratory] was across the railroad tracks; to get from one to the other during the winter I had to put on my boots, put on my coat and schlep over there,” he recalls.

But it was worth the trek. Alan's research has had applications in a staggering range of fields, from detecting arrhythmias in EKGs, to measuring the variation of the sea level over the entire planet. In fact, it was while working on a variety of signal processing projects, including work with Schlumberger – this one looking at oil well data – that Alan brought his systems and control background into play. In systems and control, using models is absolutely central. While some signal processing researchers did make use of models explicitly, that was far from universal. So, Alan put his focus on “model-based signal processing,” arguing that “without a model I don’t know how you separate information from noise…and one person’s information might be another person’s noise.” Today the use of models for information extraction from data is central in fields ranging from signal processing to machine learning.

This was part of the shift from Alan’s first field, control and systems, to his new one: signal processing. This work centered on processing and analyzing signals coming in from various sources, often in a continuous flow.

For example: When you have a stream of data coming in from a sensor, how do you tell whether the latest bit of data is unexpected or falls within the bounds of an expected range? And if it's outside of the anticipated range, is the sensor inaccurate because it is misbehaving, or is it accurately informing it you there really is something alarming out there?

At the same time, you must account for uncertainty in the external environment – such as winds buffeting a plane, or bumpy terrain that a robot's wheels must cross. And of course, aircraft and robots don't rely on data from a single sensor – they have whole banks of sensors streaming different types of data that must be stitched together seamlessly over a period of time to form the big picture.

Alan’s work, no matter what the field, (his most recent research is in machine learning), has always been impactful—as evidenced by just a few of his awards and achievements along the way. One of the earliest was in 1975, when he received the American Automatic Control Council's Donald P Eckman Award, for an outstanding young researcher in the field. In the early 1980s, he and renowned electrical engineering expert Professor Alan Oppenheim co-wrote a definitive textbook, Signals and Systems, which is still widely used today. In 2010, Alan was elected to the National Academy of Engineering.

Another experience Alan counts as formative came in 1979, when former student and MIT faculty colleague Nils Sandell started a small technology company, Alphatech, bringing in Alan and then-LIDS director Michael Athans as co-founders. The firm served mainly government clients such as the Department of Energy and Department of Defense, doing contract research and development based on the scientists' cutting-edge work.

“I learned an incredible amount at Alphatech,” Alan says. “The research was in areas like systems and control, estimation, intelligence surveillance and reconnaissance.” From a five-man team, the company grew to some 450 people over the next two and a half decades.

The experience taught him a lot about technology's place in the wider world. “When a customer comes in with a challenge, how do you creatively figure out how to use the most cutting-edge ideas to come up with something that isn't just for publication's sake, but that's going to solve that customer's problem?”

Critically, it also showed Alan and his colleagues where advances in basic theory were needed, so they could direct their research efforts. “I also learned a lot about how you build a company, and how you make it so that everybody in that company feels that they can move the dial, they can affect the future of the company, and that they're appreciated for it. That is extraordinarily important to me,” Alan adds. By 2004, when the company was acquired by BAE Systems, Alphatech had become a major provider to the federal government of advanced information technologies.

Alan has used the wisdom he’s gained over the course of his career in-part as a leader of LIDS, as well, serving as Assistant Director (1974-81), Acting Director (2007-08), Co-Director (2008-09) and Director from 2009 until he retired in 2014.

For the past two years or so, Alan has been a “professor, post-tenure”. Though he has given up a tenured position, he is still able to be principal investigator on grants, and keeps an office at LIDS, working about half time. In his spare time, Alan and his wife, artist and illustrator Susanna Natti, have remodeled her childhood home above a quarry in Gloucester, Mass., and traveled to spots as diverse as the Panama Canal and Serengeti.

As he reflects on his career, it is clear that Alan’s students are tremendous source of pride – he enjoys sitting on the doctoral committees of students whose theses he wants to learn more about, and watching former students apply their knowledge to areas as diverse as the Internet of Things and the prediction of housing prices. “I'm often surprised and delighted by the applications that I see students of mine working on,” he says.

Meanwhile, Alan continues to consult for private companies, figuring out ways to structure and build models from a mixed bag of heterogeneous data such as physical sensors and Tweets, for instance.

And one of his greatest satisfactions of the last few years, perhaps, is establishing a growing and strong statistics presence at the new Institute for Data, Systems, and Society, which applies analytical methods to societal challenges. Four new hires in machine learning are distributed between CSAIL and LIDS. For him, the line between the physical and computing worlds, “is and has to remain completely blurred”, he says.

Alan's career has been a genre-bending one with few labels, and one in which theory and application inform each other. It’s a way of thinking he’s passed on to his students, who have come from many different disciplines to find a common ground in statistical modeling and methodologies. “Statistics, Electrical Engineering, Computer Science,” he says, “None of us can figure out what field we’re in. And that’s something to celebrate.”