Meet the Faculty - Aryan Deshwal

Tell us about your journey to the University of Minnesota.

My uncle was the one who introduced me to computing and mathematics broadly. He really emphasized the importance of doing well in my academics. I got my undergraduate degree in India in applied and computational mathematics, then I moved to the United States for my PhD. I really had a great time working with my advisor, Jana Doppa. He really showed me how computing can have a broad impact on so many real world problems. I had a great time during my PhD program and he really helped me hone in on my research taste. He always used to say, “It’s better to find approximate solutions to important problems, than perfect solutions to less important ones.” It was a really supportive and encouraging environment.

When I was interviewing for faculty positions last year, I was looking for a similar environment and I felt that at the University of Minnesota. People were really nice and friendly, and it is one of the best public institutions in the country. I really like the fact that the department has people working on a broad range of topics. My research interacts with a lot of other scientific and engineering departments, so it is nice to have them within the same college. 

We would love to hear more about your research!

I work broadly on machine learning and artificial intelligence (AI), where I focus on developing new methods for challenging problems which derive from scientific and engineering applications. The overall theme for my research is AI to accelerate discovery and engineering design. Society is facing a lot of sustainability challenges from climate change to emerging healthcare problems, and solving these problems requires us to accelerate scientific discovery. Ultimately scientific discovery comes down to decision making under uncertainty, so my work focuses on developing machine learning tools to help scientists navigate their space. 

One problem that shows up in a number of scientific and engineering domains is how to explore a large and complex design space to optimize some expensive objective that we care about in a resource efficient manner. Let’s say I’m a material scientist. I would have a large space of materials and I want to explore the space to find the best material for a property I care about, like absorption of carbon dioxide from power plants. I want to do that while minimizing the resources cost of doing a costly lab experiment or other tests of that nature. A lot of different domains are dealing with a version of this problem. My research is developing machine learning methods to address this challenge, which requires probabilistic modeling and decision making policies that can be molded to the specific problem at hand. 

What do you hope to accomplish with this work? What is the real-world impact for the average person?

I think some of these applications in material, hardware and chemical design will have a big impact on addressing current societal challenges. We want to be able to have a system in place to allow designers to get the right materials they need based on the properties they care about. When we reach that level, it will have a huge impact on our ability to address societal problems. 

As another important goal of my work, I want to help my students grow and become independent researchers and professionals. My own mentors have had a significant impact on my life and I hope to pass that down to my students.

What courses are you teaching in the future? What can students expect to get out of that class?

Next semester I am planning to teach a course on AI for decision making under uncertainty. This is really the basic holy grail of AI. When we talk about modeling, finding patterns in data and making predictions, in the end they are only useful when we use them to make decisions. This course is going to cover how to make decisions under uncertainty. It will cover the key dilemma of exploration and exploitation under different settings and circumstances. Handling this dilemma is central to sequential decision making under uncertainty. 

For example, I recently went to the State Fair and I want to figure out my favorite food among a number of choices. I could continue to try the things I know I like, which would be exploitation, but I would never see anything new. I could explore new options, but I might not like everything. Or I can try a balance of the two options. The dilemma of exploration and exploitation will be the central focus of this course. I will cover the foundational principles of each as well as new methods in Bayesian optimization and reinforcement learning, and connections between them. 

What do you do outside of the classroom for fun?

I really like spending time with friends and family. I like watching movies and playing board games together. I recently went and saw Twister in theaters which was very fun. I also play some racquet sports like badminton and squash.

Do you have a favorite spot in the city?

I just moved here, but I went to the lakes around Minneapolis which was very nice. I also went to the State Fair and enjoyed trying different foods. 

Is there anything else you would like students to know about you?

I am looking for students to work with me. If they are interested, please feel free to reach out! If there are PhD students close to graduation, I was just on the academic job market and am happy to share my materials or advice around that process.