Skip to main content Skip to secondary navigation
Main content start

Madeleine Udell joins MS&E as Assistant Professor

Prof. Udell studies optimization and machine learning for large scale data analysis and control.
Professor Madeleine Udell

MS&E welcomes Madeleine Udell to our faculty!

Prof. Udell received her PhD in Computational and Mathematical Engineering from Stanford University. Most recently, she was an Assistant Professor of Operations Research and Information Engineering and a Richard and Sybil Smith Sesquicentennial Fellow at Cornell University. Now, back at Stanford, she will also be an affiliated faculty member of the Institute for Computational and Mathematical Engineering (ICME).

We caught up with Prof. Udell to ask about her current work and why she chose MS&E. Read the brief Q&A below.

Can you describe your recent work and the impact you hope it makes, both in your field and in society?

One of my biggest research topics right now is how to speed up standard optimization for machine learning by understanding the structure of data sets. Data is almost always collected for a purpose, and as a result, it has features that can reveal information about what you’re trying to predict. For example, in large data sets, many samples can be quite similar. If you figure out those similarities efficiently, you can represent the data with reduced  memory or storage.

Faster computations can help society in a few ways. Organizations can train larger and better models, and if those models are predicting how to help people in some way, then being able to make better predictions faster is beneficial. Also, training large machine learning models produces a lot of greenhouse gasses. Faster models use fewer computational resources, and in turn cause fewer emissions.

Another major research topic of mine is predictions with missing values. A lot of my earlier work deals with how to fill in missing values in order to use standard machine learning workflows. But recently, I've been thinking that imputation might not always be the right approach, because the fact that a piece of data is missing might actually be the most informative and important data. Take hemoglobin A1C tests, for example. Patients who don’t get tested are all likely to have lower hemoglobin A1C values than the average of tested patients, because doctors only order the test when a high value is suspected. If we can handle missing values well, we can do a much better job of prediction in health care.

One more particular interest of mine is electoral surveys. The standard methodology of calling people on the phone to ask questions may have been fine years ago, but in today’s world, people who pick up their phones are very different from people who don't. This is ultimately a missing data problem—we want to know about the voters who are not picking up their phones—and it's a huge problem. It's the reason why election predictions have gotten worse in the last several election cycles.

What does being a part of MS&E mean, or what makes it the right place for you?

One of the most interesting things about MS&E is its breadth. We have faculty from theoretical computer scientists all the way to qualitative researchers who go into companies and study their business processes.

I like designing algorithms, and I think of myself as a methodologist who designs new methodologies to help solve problems more effectively. It's very exciting for someone like me to be in a department this broad, because on one hand I can work with people who are going into companies and understanding what the real problems are on the ground, which will help me design appropriate methodologies. And on the other hand, I can work with theoretical computer scientists to make my methodologies more effective, scalable, and reliable through modern mathematical techniques. I'm thrilled to be able to be in a department with that sort of breadth, and one that clearly cares about impact.