Electrical and Computer Engineering curriculum prepares students to join the AI revolution

Although many of us may consider artificial intelligence (AI) to be a relatively recent development, with the unveiling of personal assistants like Apple’s Siri and large language models like ChatGPT, the term was coined in the 1950s. The field of AI has continued to grow and flourish, incorporating aspects of data mining, signal processing, and machine learning (ML).

In the last five years, the popularity and accessibility of large language models like ChatGPT have especially opened up the field of machine learning for more people to participate and use the technology.

“Machine learning and data science are permeating literally every aspect of science, industry, and government,” said Laura Balzano, associate professor of Electrical and Computer Engineering (ECE).

The ubiquity of AI-related topics has spilled over into the job market and, consequently, influenced the career aspirations of many students.

“Almost every company is looking for machine learning and data mining professionals—not just software companies like Google or Facebook,” said ECE Professor Lei Ying, when describing the new and existing courses in machine learning for ECE students.

“The students have really cried out for machine learning education, because the employers are putting value on machine learning and AI capabilities,” added Alfred Hero, the John H. Holland Distinguished University Professor of Electrical Engineering and Computer Science and the R. Jamison and Betty Williams Professor of Engineering.

ECE faculty have stepped up to the plate to address this challenge, integrating AI-related topics in classes, from the more generalist “Machine Learning for ECE” course to specialized topics like “AI-based mixed reality.”

History of AI in ECE

One of the core research areas in ECE at Michigan is Signal & Image Processing and Machine Learning (SIPML). This area and its related courses have been integrating trends in the field over the past decade. In fact, machine learning was not always included in this research area, and it has added a unique dimension to the field.

Prior to the surge of deep learning, around 2012, predictive models were created from the underlying physics and math of a system. Deep learning models demonstrated that these problems could be solved using massive amounts of existing data.

“Early on, there was a big mistrust of machine learning by the scientific community,” recalled Hero, “and then, researchers realized that you can only do so much with simple mechanistic models. It became obvious that there was a limitation to mechanistic models—that you couldn’t physically model everything, and that the whole modeling process was too slow. Data-driven models became more prominent, exploiting the fact that computation and data collection has improved tremendously over the past 30 years, enabling the data-hungry machine learning methods that are used everywhere today.”

There are examples of this transition across scientific fields: recent applications of AI have spurred innovations in protein sequencing, drug discovery, and healthcare accessibility, to mention a few. AI breakthroughs have even made their way into the humanities fields, like music.

“In music synthesis, people would spend all this time trying to model the physics of a violin. We know that when a violin plays, it’s making some combination of sinusoids; it comes from the physics,” said Jeffrey Fessler, the William L. Root Distinguished University Professor of EECS. “Now, people just record a bunch of violins and use the samples to make new violin sounds, instead of trying to model the Stradivarius sounds mathematically and physically. It would have been hard to do that 20 or 30 years ago, because they didn’t have the amount of data storage or computing power needed to train these models—so we used the mathematical models that could be run on lower-speed processors.”

ECE faculty have played an essential role in developing the hardware to enable larger data storage and faster processing, and they continue to advance hardware for AI and machine learning applications. But they are also playing a key role in developing machine learning algorithms and ensuring that students have the skills to use machine learning and AI in their careers, whether in research or industry.

Teaching fundamentals and tech trends

Foundational courses in machine learning have been taught at Michigan for at least the last decade. Undergraduate students can take EECS 453: “Principles of Machine Learning” and graduate students can opt for EECS 553: “Machine Learning for ECE.”

“It was clear that machine learning was something that our students were well-prepared to study, master, and go out and use,” said ECE Professor Clayton Scott, who regularly teaches EECS 553. “Each time I teach ‘Machine Learning for ECE,’ I add one or two new lectures to incorporate more deep learning and more modern machine learning to keep up with the times and the demand.”

EECS 453 and 553 are especially tailored for ECE students. For example, they include more mathematical background and principles compared to their Computer Science and Engineering counterparts (EECS 445 and EECS 545). 

In addition to EECS 453 and 553, ECE provides a range of courses that either overtly, or as part of the syllabus, provide ECE students with the knowledge to understand AI/ML. Several have been discussed in more depth in a previous story, “Teaching Machine Learning in ECE.” 

Here is a list of additional courses, updated from the previous story:

• ENGN 100: “Computational ML for Scientists and Engineers”
• EECS 298: “Applied Computational Machine Learning”
• EECS 452: “Digital Signal Processing Design Laboratory”
• EECS 505: “Computational Data Science and Machine Learning”
• EECS 551: “Matrix Methods for Signal Processing, Data Analysis and Machine Learning”
• ECE 556: “Image Processing”
• EECS 598: “Statistical Theory of ML Limits”
• ECE 598: “Large Language Models”
• ECE 598: “Randomized Numerical Linear Algebra for ML”
• ECE 598: “Statistical Learning Theory”
• EECS 559: “Optimization Methods for Signal & Image Processing and Machine Learning”
• ECE 602: “Reinforcement Learning Theory”
• EECS 605: “Data Science and Machine Learning Design Laboratory”

Topping that list was a course for first year students who haven’t even declared a major, ENGN 100, to be offered for the first time in Fall 2025.

“My goal for my students is to get them to understand that there is some math behind these AI models that needs to be harnessed properly to take advantage of all its functions, and then to give them a feeling of what it means to actually tune an AI model,” said Raj Nadakuditi, associate professor of ECE. “That’s the journey I want to lead them through—how to frame, evaluate, and continuously improve these learning systems, mimicking the technical advances that we teach them in subsequent courses.”

“Machine learning is such a vast field that you can’t cover it in one or even a handful of courses,” Scott said. “We try to impart the ability for students to to teach themselves—we give them enough exposure to get them started and they take it from there.”

Four new courses that apply AI to achieve specific goals

In addition to the courses listed above, many professors have recently developed, or are in the process of developing, new courses to address particular applications of AI. Here, we highlight four of these courses.

Computational Power Systems (ECE 598)

Vladimir Dvorkin has designed his course, “Computational Power Systems,” to cover the use of AI and machine learning to aid power grid operations and electricity markets.

“In the past, we had simple electricity demands. Residential consumers had very predictable schedules, and that made electricity consumption extremely predictable,” said Dvorkin, assistant professor of ECE, “But now there are so many other resources in the system, like data centers that work through the day and night, or renewable energy that depends on the weather, so there is a lot of uncertainty and variability in the system. Consequently, our decision making needs to be fast; this course is about the algorithms that support decision making in rapidly evolving power grids, making them reliable and economic by means of AI and optimization.” 

The course has been popular, drawing in students from across Michigan Engineering, and even from the College of Literature, Science, and the Arts. The importance of power systems to humans’ wellbeing—and its increasing complexity—leads many graduate students to pursue research on its operation and optimization.

These quotes from students in three different departments provide unique insights into the course:

“The entire class revolves around optimization methods, which I regard to be the fundamental of AI. I would highly encourage anybody interested in AI and optimization to take this class because it not only introduces you to the mathematical background, it also puts it into application context,” said Xinwei Liu, ECE PhD student. “Power energy will be one of the most important problems in our century and applying our knowledge in this field is crucial for the future of our generation.”

“Machine learning has many powerful tools with a plethora of uses in the power systems space; in this course, we get a good introduction to the current suite of applications,” said Megan Jones, PhD student in the School for Environment and Sustainability. “I will utilize the techniques taught in this class for approaching common power systems modeling problems in my research on capacity expansion and reliability modelling of the grid.”

“I took this course because I believe that many important theoretical questions are driven by real-world needs. Power systems, in particular, pose complex and pressing challenges—from uncertainty in renewable generation to large-scale, distributed decision-making. Modern power systems demand smarter, more adaptive solutions,” said Qingyuan Xu, PhD student in Industrial and Operations Engineering. “This course gave me the opportunity to explore those challenges up close and think critically about how rigorous optimization frameworks can support real infrastructure and policy decisions. It also pushed me to think more intentionally about the kinds of theoretical tools we develop, and whether they’re truly suited to the systems we aim to support.”

Artificial Intelligence in Biomedicine (ECE 598)

Liyue Shen’s “Artificial Intelligence in Biomedicine” course highlights uses for AI in human healthcare. Algorithms can help scientists and doctors with tasks such as identifying anomalies in images like MRIs, x-rays, photos of skin, and microscopic views of cells. Ultimately, AI in biomedical settings has the potential to help diagnose disease, develop vaccines, and more.

“The students try to think about more interesting, innovative ideas for doing the image reconstruction more reliably and efficiently,” said Shen, assistant professor of ECE. “We have so many unique challenges with the many different and complicated biomedical data types—like CT, MRI, X-rays, ECG, EKG, genomics, and electronic health records. The students use data in collaboration with Michigan Medicine to build rare and very valuable multimodal biomedical datasets.”

Shen’s students get to learn about new research topics and methods that they haven’t learned about in previous image processing classes and apply deep learning techniques. It gives them a chance to apply up-and-coming methods to topics that have the potential to improve and save human lives. For example, Shen and Fessler have collaborated on models that improve the efficiency and quality of low-dose CT scans to improve medical imaging without dangerous side effects of radiation.

“People are always wondering, ‘Can we just use the deep learning approach to address this problem?’ I feel like the students do have more interest to understand and learn more about how AI and machine learning have been used in the cutting edge research of this field,” she said.

AI-based Mixed Reality (ECE 598)

Jiasi Chen has developed a class that teaches students to apply AI algorithms to virtual, augmented, and extended reality through existing, out-of-the-box headsets, laptops, or phones. Pokémon GO is one example of a mixed reality application running on a phone.

In “AI-based Mixed Reality,” students use machine learning in both the software and to support the underlying hardware of mixed reality implementations. They gain skills in using existing machine learning models and integrating them with a headset, which comes with its own technical challenges.

“AI is more of a research topic, so if you go to the keynote at an academic conference or talk to your collaborators, they’re all really interested in AI and how it can be applied to mixed reality,” said Chen, associate professor of ECE. “I looked at other mixed reality courses on campus, but they’re more focused on applications and software engineering, so this is more a mix of practical things you can actually apply and research projects.”

One example of a student project is a live transcription product, where a headset will automatically translate speech in a different language and display it as a speech bubble, like in a video game. Another project trains a model to recognize hand gestures used in scuba diving.

“As a seminar course, it’s also about reading and understanding ideas; there’s no textbook on this,” said Chen.

Computer Engineering for and with AI/ML (EECS 498)

Robert Dick is unveiling a course in Winter 2026 to teach students how to design computer systems to support AI/ML applications, and using AI/ML approaches for computer system design.

It will include four major topics: an introduction to AI/ML fundamentals; a review of the practical uses of AI/ML systems, including big data and small data problems; practical approaches to designing computer systems to efficiently host AI/ML applications, and using AI/ML for computer system design. The last topic will include LLMs for software generation and system-level design automation.

In this project-based course, students will also learn about the basics of deep neural networks, common problems that students may encounter when using them, and practical approaches to designing software and hardware to host and execute AI and machine learning applications.

“AI/ML based applications and solutions are very common. Computer engineers will benefit professionally and academically from understanding them,” said Dick, professor of ECE.

Is it AI or machine learning or signal processing?

In ECE, many professors are hesitant to call the topics that they teach in their course—or research in their groups—AI, finding the term “machine learning” to be more accurate. But what is the difference? Individual answers differ.

Chen describes machine learning as a topic under the larger umbrella of AI. “I would say they go from most to least general, so AI is the whole space, ML is like specific types of AI—deep learning, transformers, large language models—and I would say signal processing is a technique that can be used to do machine learning along with many other things,” she explained.

Hero makes a distinction between the inference, or application, of the models, and their training or learning process. “AI has come to designate the software and its application to a particular set of data for a particular purpose, whereas machine learning is now the term for the art of the way that AI works—studying the fundamental principles, mathematically describing the process, training the system using existing theory, and then evaluating how well the system might perform in the presence of different challenging environments,” he said.

Scott explained that part of the difference is in the inspiration and the goal for each type of work. “There’s a view by some that 99% of what people call AI these days is just machine learning,” he said, “but at the same time, with AI, there’s a goal of actually building up to mimic or surpass human intelligence, not just solve concrete scientific problems or advance fundamentals, like we do with machine learning.”

Over time, the words used to describe certain research topics may be changing as the work advances. “Now, everyone who was trained in signal processing is doing machine learning,” Scott said—a sentiment that was echoed by Hero.

Either way, the topics that ECE students learn in their courses will be applicable to the AI they will use and develop in their careers, whether in research or industry.

“As AI has become part of the common vernacular, I have reminded students that the things that we’re doing in that class [ENGN 100], even though I’m not calling it AI, are techniques that they will see again as they learn more about machine learning and AI,” said Fessler.