Epod Episode 6: Jeff Linderoth on Design Optimization

Listen to Episode 6:

On this Episode:

On this episode, Susan Ottmann talks with Dr. Jeff Linderoth from the Department of Industrial Systems and Engineering about engineering data analytics and design optimization. Jeff discusses the fundamentals of optimization and how it fits into a variety of applications with global implications.

Our Guest:

Jeff Linderoth is a professor and department chair of the Department of Industrial Systems and Engineering at UW-Madison and a fellow with the Wisconsin Institute for Discovery. His research focuses on modeling and solving real-world, large-scale optimization problems. Specific research areas within optimization include integer programming used for modeling yes/no decisions, and stochastic programming useful for decision making under uncertainty. Jeff’s research places a particular emphasis on developing high-performance, distributed optimization algorithms, and software. He has an MS in  Operations Research and a PhD in Industrial Engineering from the Georgia Institute of Technology. In his free time, he enjoys golf, tennis, cooking, and chess.


Subscribe to Podcast



Justin Kyle Bush: Welcome to Epod, a podcast from UW-Madison’s College of Engineering’s Office of Interdisciplinary Professional Programs. These podcasts are focused on big ideas and engineering and the people behind them. My name is Justin Kyle Bush, and I’m your host. On today’s episode, Susan Ottmann talks to Dr. Jeff Linderoth from the Department of Industrial Systems and Engineering about engineering data analytics and design optimization. Jeff discusses the fundamentals of optimization and how it fits into a variety of applications with global implications. Take it away, Susan.

Susan Ottmann: Welcome Jeff.

Jeff Linderoth: Well, it’s fabulous to be here. Thank you, Susan.

Susan Ottmann: Before we start, Jeff, please give us your view of the Engineering Data Analytics program at UW-Madison.

Jeff Linderoth: Well, what I think really sets the Engineering Data Analytics program apart is its engineering focus. If you look around there are many Master of Data Science or Master of Analytics programs in the marketplace today. And those are great and they are going to teach you tools and concepts to effectively analyze data. The MEDA program here at UW-Madison actually goes beyond that and shows you how to apply that data to solve engineering problems. I mean, the classes on engineering design systems, reliability, simulation modeling, just to name a few, really give the program an engineering focus that I believe sets it apart.

Susan Ottmann: Let’s dive a bit deeper and talk about design optimization and why this is important for engineers focused on data.

Jeff Linderoth: Well first, optimization is a mathematical discipline that’s all about finding the best solution to a problem, given constraints or scarce resources. Second, and I’m probably not going to surprise anyone with this observation, is that there really is a data revolution going on today.

More and more companies are marrying these two concepts of optimization and data. In fact, it’s led to a new term of art called data-driven optimization. And this is where the model of a process or a system is deduced or inferred solely from data or historical observations. Even more exciting is when this data can help inform an engineering-based model, maybe a math-based model,

and this is done via estimation of key parameters in the system. Once the parameters are estimated with data, that design and design optimization process can begin. So, I mean, if you’re an engineer focused on data, doing optimization, design optimization, really is one of the most important tools that you can have in your toolbox.

Susan Ottmann: I find the whole area of design optimization. Fascinating. We’ve got optimization courses available in mechanical and industrial systems. How do you see the applications differing across the courses in your department and in mechanical engineering?

Jeff Linderoth: Yeah. Well, one of the differences is that some of the courses anyway, in industrial and systems engineering, the courses are a little bit more focused on theory and algorithms.

But really, I believe the fundamental difference between optimization and mechanical engineering and an industrial and systems engineering is on the class of applications problems. In fact, you know, we added the “systems” name to our department about 15 years ago to acknowledge the important role that systems play in our discipline.

Maybe I can best, uh, answer your question with an example. So imagine two engineers working at the same factory. Mechanical engineers would potentially be employed analyzing, you know, optimizing a specific process within the factory, maybe a machining process to minimize component wear.

Industrial and systems engineers working at that same factory would be using optimization to lay out the shop floor holistically to maximize product throughput. Now, both of these are extremely important, both the finer level of detail to get the individual processes right and efficient, but then also how to connect these processes together in the most efficient way.

Susan Ottmann: I love process, but I’m also a mechanical engineer and that makes a lot of sense to me. What about those electrical engineers out there who are enrolled in our degree program? How do you see design optimization applying in their work?

Jeff Linderoth: Oh man, optimization is everywhere. I have many problems in signal processing can be converted to optimization problems. Actually one of the most classical applications of design optimization is in tuning and in the optimal design of circuits for performance and reliability. I myself actually have done optimization for the design and operation of electrical power grids, and I’ve used optimization in VLSI chip design.

I mean, honestly, there are optimization problems and every discipline of engineering. The world needs all types of engineers and, in my opinion anyway, all engineers should know about optimization.

Susan Ottmann: One of your research areas is global optimization. Please tell us more about your research and interest in this area.

Jeff Linderoth: Well, if I had to sum up my research, it would be that I like to try and solve the hardest classes of optimization problems and global optimization is one of these hardest classes. So, optimization is like trying to find the top of a mountain. And the design landscape can be very hilly. Many optimization methods look locally. So they might look at where they’re standing in the design space and they look around at the neighboring designs. And if they don’t see any, any close neighboring design, that’s say, better than they are. Then they declare themselves to be, “I am the best design. I am optimal.”

However, way out in the distance beyond their neighborhood, there may be a higher peak, there may be a better design. And global optimization is all about studying the mathematical properties of the optimization problem and designing algorithms that can find that higher peak. Not only that, but it can prove mathematically that no, a better solution exists—there doesn’t exist a higher peak.

So I also love it when my algorithms and analysis can really make a difference. So I love teaming with engineers and scientists to help solve the optimization problems in their domain that will make a difference.

Susan Ottmann: Jeff, I know you’ll be off on sabbatical next year. Can you give us a sneak preview into the research you’ll be pursuing?

Jeff Linderoth: Uh, well, besides trying to lower my golf handicap for one…actually, no, I have three really exciting new projects that I’m just sinking my teeth into that I think will take a lot of my energy during my sabbatical. The first one is a machine learning application. This is a little bit like the very well-known Netflix challenge where we are given us a subset of user ratings for movies and the goal or the task is to try to predict or complete the ratings for movies that the user hasn’t seen. However, in this application, a special twist is that the users may come from completely different classes of individuals. So it’s a double challenge of first trying to infer from the data which class a user falls into, and then given that, to try to complete his or her ratings of the movies. And actually, the actual application of this is actually in drug design, trying to predict drug disease, interactions, not user, uh, movie ratings.

A second application I’ve just started is in computational, or actually, evolutionary biology. So here though, we’re given a database of potential chemical reactions and the optimization problem is to find a small so-called auto-catalytic network, which is just a series or a set of reactions that could be self-sustaining to produce organic compounds that form the basis of life.

And then a third application I’ve just starting is on the optimal deployment of Naval assets to mitigate adversary threats. So given the potential, I’ve just started working on this project, given a potential red layout and potential red threats, we have to deploy our blue assets and in a way, in an optimal way, to mitigate those threats. So those are just three—and those are all optimization problems. So those are just three of the very diverse set of optimization, or applications, of optimization.

Susan Ottmann: All I can say is wow. I find your research so interesting and believe that you’re helping to move the world forward in the area of data, data, analytics, and optimization. And just those three projects are so diverse that optimization plays a role in so many parts of daily life and engineering. Well, thank you so much for your time today. We appreciate you talking to us and informing us. Very interesting topic.

Justin Kyle Bush: Thank you for listening to Epod. For more episodes, visit www.interpro.wisc.edu/podcasts. And if you enjoyed this, don’t forget to subscribe, rate, review, and share.