Interview with Kevin Luo, May 2024 Master’s Prize Winner

Congratulations to Kevin Luo, the May 2024 recipient (along with William Nickols) of the Department of Statistics Concurrent Master’s Prize! The prize is awarded annually to up to two graduating students having completed the Concurrent Master’s program in Statistics who have the best overall performance (as indicated by coursework results), have demonstrated achievements in Statistics outside of coursework, and have contributed significantly to the department.

To learn more about Kevin’s inspiration, thesis experience, and sense of community within the department, we spoke with him. Highlights from our conversation are edited and excerpted below.

What sparked your initial interest in math or statistics?

Luo: My first exposure to statistics was when I completed a science project in eighth grade. The lakes in my neighborhood in Germantown, Tennessee had turned increasingly green over the years due to fertilizer runoff and, as a result, most of the fish disappeared (when I was little, I could reach my hand in the water and grab a fish!). My mom, who works in bioinformatics, suggested doing toxicity testing on the water and, in the process, introduced me to hypothesis testing. While I didn’t fully understand hypothesis testing at the time, I started to frame statistics as a useful tool that lets you control for error, making it feel less rigid than the math I knew. This early impression influenced my decision to study statistics when I started attending university.

What are your initial memories of the Stats Department?

Luo: My first experience with the stats department was taking Stat 110 Probability during my freshman fall in 2020, a year that was entirely online due to COVID. Based on my schedule, I ended up in a 10:30 p.m. section for the course! It turned out to be an amazing section, run by Yash Nair (’22 AB alum) and Junu Lee (’22 AB alum), who later graduated and went on to top graduate programs. They were a big influence on my decision to pursue statistics; their section also helped me to have fun and meet other students during an isolating time.

What challenges did you face transitioning to Harvard and the Master's program?

Luo: When I arrived at Harvard, I experienced a lot of imposter syndrome because I looked at others’ accomplishments in high school and did not feel as prepared. Not wanting to seem “dumb” initially made me hesitate to study with other students. COVID also didn’t help because all of our classes were on Zoom and there were no gatherings in the dining hall, making it hard to form friendships. Gradually, my fears faded away when I realized that in college, it no longer matters what you did in high school because you are free to choose a new path and go at your own pace. As a sophomore on campus in person, I made a bunch of friends, which also helped me to pursue my interests without worrying about others’ judgment.

The transition to taking grad level courses was pretty smooth because there is a strong culture at Harvard of undergrads taking graduate courses. Going from undergraduate to graduate courses was also comfortable because Professor Joe Blitzstein teaches both Stat 110 and the first graduate level class Stat 210 Probability I. While the content of the two courses is different and Stat 210 is more challenging, they follow the same style and structure.

What motivated you to pursue the concurrent master’s program?

Luo: The best advice that I received was from Kim Nguyen (’21 AM alum), who said, “Take the classes you’re interested in, and then in senior year, figure out which degree you can graduate with by taking the fewest additional classes.” I followed her advice and ended up not only being a statistics concentrator but also receiving a master’s degree, which is a pretty good deal!

A specific course to highlight would be Stat 217 [Topics in High-Dimensional Statistics: Methods from Statistical Physics] – it changed my life in my junior year! Taught by Professor Subhabrata Sen, the course introduced me to statistical physics. Two contrasting ideas, which are really two sides of the same coin, were very cool to me. The first was universality, which roughly means that some large system behaviors are not sensitive to small details. One example of universality is the central limit theorem. The second was the concept of phase transitions and order parameters, which say that, on the other hand, there are certain parameters that entirely determine the large system behavior, and this behavior can change sharply when these parameters are slightly perturbed.

How did you start collaborating with Professor Pragya Sur on your thesis and how did you select your thesis topic?

Luo: With a broad interest in working on a project related to machine learning and theoretical statistics, I began skimming research papers of faculty and talking to them. I chose to work with Professor Pragya Sur because we discussed a project that I thought would help me grow and learn the most.

The original goal of my thesis was totally different from the final product. I started out working with Professor Pragya Sur and her student Yufan Li (’25 PhD alum) on random feature models. I was trying to prove something about the asymptotic mean squared error, but I failed because I didn’t understand the techniques well enough. I remember having a tough time with the project over winter break, but it was a good first lesson in the research process and how to move forward after failure. I realized that you can’t just expect a solution to unravel in front of you; instead, you need to work on small pieces of the problem, even if it feels far from helping you solve the whole problem. It was a humbling experience.

We pivoted to a different, simpler problem: high-dimensional ridge regression under dependent data settings. Ridge regression is related to linear regression but is used to combat overfitting by adding a penalty to the cost function for complex solutions. Most existing high-dimensional ridge regression analysis assumes datapoints are independent and identically distributed (IID), but in real-world applications like medicine or finance, this is often not the case. For example, medical data for a family, such as weight and height, are likely to be correlated, not independent variables. Using models that allow for data dependence, we explored what happens when data points are dependent and whether conventional tools still work. We found that many behaviors remained similar, but common cross-validation methods broke down in this setting with this data, so we proposed a new approach.

What did you value most about your experience in the department and at Harvard? If you had to choose one word to describe it, what would it be?

Luo: I most valued the opportunity to have one-on-one conversations and work closely with faculty. I never imagined this kind of relationship with faculty when I was in high school. My advisor, Pragya Sur, helped me to understand the landscape of statistical research, the problems researchers care about, and my research interests. When Pragya advised me on my thesis, she never seemed surprised about anything—she always understood the problems extremely well and had great intuition.

Related to my experience with Pragya and other faculty, I would use the word “defining” to sum up my Harvard experience. When I entered college, I had no clear idea of what I wanted to pursue (I even wrote my college essay about this!). In my first few years at Harvard, I dabbled a lot in different subjects, and this provided me with a strong understanding of my interests.

What are you excited to pursue? Describe some of your career, academic, and personal aspirations and plans.

Luo: This summer, I’m excited to revisit the research project that initially failed, the one on random feature models. During the semester, there were some nights when I could not fall asleep because I was thinking about the problem and what I would do differently! I now realize what to do, so I would like to start working on the problem again. I’m also looking forward to spending time with some of my high school friends, whom I’ve stayed very close with especially because of my COVID freshman year.

In the fall, I’m excited to start as an algorithm developer at Hudson River Trading. My work will likely center on building systems for stock trading strategies.