The Professor Behind 5G Wireless Innovations, Data-driven Recommendations and MRIs

nikos_800.jpg

Throughout his career, Nikolaos Sidiropoulos, the Louis T. Rader Professor of Electrical and Computer Engineering at the University of Virginia, has made key contributions to signal processing and communications technologies that we now consider ubiquitous, such as magnetic resonance imaging – what we know as MRIs – ad recommender systems, and 4G and 5G wireless data transmission. He also contributed to high-resolution hyperspectral imaging, a technology that can acquire extremely detailed information for things like environmental analysis to battle climate change and many other applications.

The Institute of Electrical and Electronics Engineers Signal Processing Society honored Sidiropoulos with the 2022 Claude Shannon-Harry Nyquist Technical Achievement Award, a lifetime achievement award “given for outstanding technical contributions to theory and/or practice in technical areas within the scope of the society, as demonstrated by publications, patents or recognized impact on the field.” The citation noted Sidiropoulos’ “exemplary contributions to tensor decomposition, beamforming and spectral analysis.”

Scott T. Acton, professor and chair of the Charles L. Brown Department of Electrical and Computer Engineering, called the Shannon-Nyquist award “the Turing Award of signal processing.” The Turing Award is the highest distinction in computer science and is colloquially referred to as the “Nobel Prize of Computing.”

The past winners are the titans of the field, Acton said, including Al Oppenheim, who co-authored some of the most widely used signal processing textbooks, and Sidney Burrus, who helped develop several concepts underpinning digital signal processing.

“Nikos continues to innovate and drive foundational theory in our field,” Acton said. “His work in tensors, for example, is the basis for industrial products such as recommender systems. These systems choose your next Netflix series or what ads accompany your Gmail.

“Nikos also pioneered techniques in beamforming, which enable your cell phone to reliably connect and communicate. We are lucky to have Nikos in Charlottesville, Virginia!”

Sidiropoulos describes his work as “I use applied mathematics that I develop to solve different engineering problems,” Sidiropoulos said to describe his work.

Tensor decomposition, for example, is taking multiple complex data inputs and mathematically reconfiguring them into something simpler and usable. As Acton mentioned, one usage allows streaming or digital services to pull together multiple data points in a formula to generate recommended content or ads for consumers.

Color is another good example. Digitally, a color image is rendered with three input matrices – red, green and blue – which together form a tensor. Sidiropoulos’ work helps computers fuse multiple image inputs – multiple tensors – to recreate detailed hyperspectral images of the earth, and high-resolution magnetic resonance images at faster acquisition speeds. 

“Processing color as a ‘third order’ tensor allows us to have very high-resolution reconstruction after fusing multispectral and hyperspectral images,” he said. “That is one of the applications of my research.”

The award citation also recognized Sidiropoulos’ critical contributions to beamforming, a signal-processing technique for focusing an array of antennas in specific directions. His work had a significant impact on the development of Evolved Multimedia Broadcast and Multicast Services, or eMBMS, which enabled more efficient dissemination of streaming content and software updates over LTE and 5G networks.

“My work is basically bringing in these tensor factorization tools to bear on important problems in machine learning and wireless communications,” Sidiropoulos said. “The best way I describe myself is as a modeler. I think about engineering problems and model them using appropriate mathematics.”

In addition to the lifetime achievement award, the IEEE SPS simultaneously recognized Sidiropoulos with a Best Paper award for a paper he co-authored, “Learning to Optimize Deep Neural Networks for Interference Management,” and the Donald G. Fink Overview Paper Award for “Tensor Decomposition for Signal Processing and Machine Learning,” which he also co-authored. “It was,” he joked,  “a triple-pointer.”

Sidiropoulos knew as a teen growing up in Greece that he wanted to do this kind of work.

“When I was about 15 years old, the father of a friend of mine was an electronics engineer and he gave me a small FM transmitter as a gift,” Sidiropoulos said. “I got hooked on building transmitters and experimenting with them. From that moment, I knew I wanted to become an electrical engineer, though I could not foresee the direction I would take.”

That ambition eventually took Sidiropoulos to the United States, where he earned his Ph.D. from the University of Maryland, College Park in 1992. In 1997, he joined the UVA faculty as an assistant professor for three years before leaving for appointments at the University of Minnesota, Minneapolis and, for 10 years, the Technical University of Crete in Greece. He returned to UVA in 2017 to chair the Charles L. Brown Department of Electrical and Computer Engineering.

“Working in several different places always keeps you on the edge and being creative,” Sidiropoulos said of his career moves.

Throughout, Sidiropoulos has loved teaching – especially the moment he sees his students connect the dots and grasp something new.

“I enjoy the opportunity to excite my students, getting them to think out of the box, to smile, to say ‘ah-ha!’” he said. “That ah-ha moment is what you live for, both in teaching and in research.”

He compared good teaching to good theater.

“Your job is to excite and motivate the students to delve deeper,” he said. “Anyone can read the book. How do you motivate them to delve deeper and become independent in pursuit of additional knowledge?”

For Sidiropoulos, the pursuit never stops. He is now focused on the next evolution of machine learning, using what he has learned about tensor composition to solve problems in the rapidly growing world of artificial intelligence.

AI, Sidiropoulos said, is currently marked by huge, complex models like the one behind ChatGPT, a chatbot that can rapidly pull together information to write seemingly intelligent, complex responses. The complex models work, but they are difficult to replicate and difficult to use without vast amounts of data. Sidiropoulos believes his work in tensors can help pare down models for quick, effective usage.

“There are many applications where you do not have a lot of data available, such as when you are trying to decide if a pharmaceutical drug is working and you have limited clinical trial data,” he said. “In those cases – with limited data but high social impact – our current models do not work well, and we do not necessarily understand them well. I am interested in building more principled and parsimonious models that would allow us to make good predictions with limited data and make predictions that are more understandable.”

In communications, he is also looking at extending the range of devices that make up the “internet of things” – all of the physical and digital devices around us that connect through the internet to operate, such as turning on lights in our homes, playing music, allowing us to check on our pets while we are away and so much more.

“I am interested in applying our tools to enhance the internet of things and other wireless communications,” Sidiropoulos said.

For example, could someone’s smartwatch or virtual reality headset and phone be modified to operate on the same frequency, so that they can work together at greater distances and faster speeds? Sidiropoulos thinks so, and he has filed a patent with UVA’s Licensing and Ventures Group to see if he can bring that idea to life. 

“That is one thing I love about my work,” he said. “We have a lab, with advanced software and broadcast equipment, to test and verify our assumptions. We are not only theorists, we are also building prototypes to check our ideas here.”

With discipline and luck, those discoveries might be part of his next lifetime achievement award.