Artificial intelligence models are a marvel of engineering, and, says Omar Chehab, reflect an engineer’s mindset.
“While building AI, people have experimented with many things. They have some intuition for why it works,” says Chehab, who joined the University of Wisconsin-Madison Department of Electrical and Computer Engineering as an assistant professor in fall 2026. “In the end, you get this very big machine that works well, but is a patchwork of tricks that we don’t fully understand.”
In his new lab at UW-Madison, Chehab aims to tackle AI problems with a mathematician’s or computer scientist’s perspective before rebuilding it like a well-informed engineer. “The idea is that we will pretend we don’t know at all why AI works,” he says. “Then we will break it down into LEGO pieces, understand the different pieces and how they fit together. Finally, we can reassemble them in different ways to build new AI algorithms.”
These new algorithms, he hopes, will be more understandable, powerful and use far fewer resources than the current generations of AI.
Chehab, who grew up in France, received a master’s degree in applied maths and engineering science from ENSTA Paris and another master’s degree in applied maths for computer vision and machine learning from ENS Paris-Saclay.
But it was during a sabbatical year in his studies that he found his true passion: data science. “For me, prior to that, data science was sort of a mystical term that was just appearing in the field. But then I learned more about it and it appealed to me” he says. “It was sort of an engineering or math way to formalize what information means and how to process it, which is something that is both intimately human and something that is now at the center of machine understanding.”
This led him to a PhD program at Inria, the French national research institute for digital science and technology at the University of Paris-Saclay. There, he studied machine learning algorithms, their statistical foundations as well as their applications to brain imaging, giving him insight into both the theory and applications of the emerging discipline. He continued this line of work in his postdoctoral positions at ENSAE Paris and at Carnegie Mellon University.
With the launch of his own lab at UW-Madison, Chehab aims to push further this interdisciplinary line of research that he has embraced from the start. His research ranges from the theoretical foundations of generative AI algorithms such as diffusion models, to learning causal networks in the brain and, more recently, generating plausible molecular configurations. “I’d very much like for my lab to emulate the research training I have had in my own PhD and onward, which is this mix of theory and practice,” he says. “But not in a generic sense. I want students to confront a specific data type, like molecular data or brain imaging data, and work with a specialist who really guides them through the intricacies of that specific data type. Then, on the theoretical side, we look at the mathematics behind the algorithms.”
Chehab says the collaborative environment at UW-Madison is ideal for this work, which will bring together his students and world experts in various fields to work on a common project. “I want to run an interdisciplinary lab that asks mathematical questions, but also wants to make things work at scale for real-world data,” he says. “That’s a very good place to be.”
Photo of Omar Chehab by Joel Hallberg