Machine Learning & Signal Processing Instrumental for AI Applications

Electrical and Computer Engineering Professor Describes How Signal Processing is at the Core of AI Technology

The emerging application of artificial intelligence (AI) to a diverse range of fields has positioned it as a valuable research tool. Stony Brook’s Institute for AI-Driven Discovery and Innovation hosts faculty from a wide variety of disciplines who are advancing machine learning research.

Part three of our AI Researcher Profile series brings us to the Department of Electrical and Computer Engineering, (ECE) in the College of Engineering and Applied Sciences for a conversation with Petar Djuric, Professor and Chair of ECE about his theory and methods research and its application to machine learning.

AI Institute: What is your connection to AI and AI-related research?
Petar Djuric: One of the pillars of AI is machine learning (ML), and at its core is signal processing. My bachelor’s thesis at the University of Belgrade was on ML methods for classification of radio signals. My first job was in the largest research institute in Belgrade, and there I was involved in a large-scale AI project, which was based on the state-of-the-art methods. Ever since, in one way or the other, I have worked on theory and methods that find direct application to ML.

AI: How long have you worked with AI and machine learning? Can you comment on your background and experience with these technologies?
PD: My first research task was in the area of machine learning. Before coming to the U.S. for my PhD, I was involved in research related to building a large AI system. In my second year after getting my BS degree, the Institute where I worked organized a summer workshop on AI. Speakers at the workshop included Patrick Winston from MIT and Allen Newell from Carnegie Mellon, both early contributors to the field. Thus, from my formative years as a researcher, I was exposed to AI, extensively. Then, in the US, I did my PhD in the general area of statistical signal processing, which remains my main area of research. With my background in statistical signal processing, I can readily apply my knowledge to solve hardcore problems of ML, ranging from first principles to a rigorous and theoretically sound framework. Most importantly, the theory that I have learned allows me to work on novel ML methods both from theoretical and practical points of view.

AI: How do AI and machine learning fit into the Electrical and Computer Engineering areas of study and research?
PD: One might argue that the most complete coverage of AI in an academic department can be in ECE. We should keep in mind that AI is a broad field that involves several general areas of knowledge. One of these areas is sensor technology. AI needs sensors to acquire signals with information, which are then used in a meaningful way in AI systems. Electrical engineering plays a central role in this domain.

Then there is hardware for AI. For instance, deep neural networks that have recently started surpassing humans in some cognitive tasks are computationally intensive and require hardware platforms with increased efficiencies. Research on AI hardware that explores new devices and architectures is critical and currently a very active area of research, and it is most likely to be carried out in ECE departments.

ML, of course, is instrumental for AI, and since signal processing is at the root of most ML methods/algorithms, it is taught in most ECE departments. One more crucial component of AI systems is software. Again, ECE departments have courses that address software for AI. Furthermore, control is an essential component of AI systems, and research as well as courses in control are standard in ECE departments. Finally, one needs to integrate all this into an AI system, and system integration, too, is instructed and researched in ECE departments.

AI: Do you think that students in your field should also take AI-related classes and expose themselves to more AI-related technology?
PD: The answer is a resounding yes and this has already been happening. It is our vision that the ECE department will have an increasing number of courses related to AI. Some of them will be modified to address current advances in the field, and others will be completely new.

AI: What do you see as future applications of AI and AI-related technology as it applies to your fields of research?
PD: I am personally interested in developing ML methods for medical applications (and by extension future AI systems). Currently, I am part of a team that works on analyzing fetal heart rate tracings and uterine activity signals in the second stage of labor with the objective of assisting obstetricians in making better decisions when a fetus may be compromised. Another big area of interest is contributing toward improved understanding of the most complex system we know of: the human brain. Once scientists and engineers have a much better grasp of how intelligence is created in the human brain, they may be able to engineer similar or even more advanced intelligence.

AI: Do you have any additional comments about AI and AI-related technologies as per your experience?
PD: We live in very exciting times of AI science and engineering. In the near future, we will see increased applications of AI, from manufacturing to medicine (personalized medicine, in particular). One important area of AI that is still quite immature is reasoning. In the years to come, it will most likely see significant advances.

AI: Thank you for your time, Professor Djuric.

About the Researcher

Petar M. Djurić is a SUNY Distinguished Professor and Chair of the Department of Electrical and Computer Engineering and faculty member of the Institute for AI-Driven Discovery and Innovation at Stony Brook University. He is also a Fellow of the Institute of Electrical and Electronics Engineers (IEEE) and a Fellow of the European Association for Signal Processing EURASIP.

His research interests include:

  • Sequential signal processing
  • Machine learning
  • Monte Carlo methods in signal processing
  • Signal and information processing over networks
  • Complex systems 
  • Analysis of fetal heart rates and uterine activity signals
  • Brain signals 
  • Signal processing for molecular biology
  • Model selection
  • Detection theory
  • Parameter estimation
  • Non-parametric Bayesian theory
  • Signal processing for mobile communications
  • Prediction of rare events
  • Spectrum analysis
  • Causality