# Postdoctoral Fellow Kelly W. Zhang is Interviewed for Women in Statistics and Data Science Series July 28, 2023 Earlier this year, we launched a series featuring women scholars in our department in honor of the inaugural International Day of Women in Statistics and Data Science, a day introduced by the [Caucus for Women in Statistics](https://cwstat.org/), the [Portuguese Statistical Society](https://www.spestatistica.pt/en), and the [American Statistical Association](https://www.amstat.org/) to celebrate the research contributions of women statisticians and data scientists around the world. Our interviews with our faculty, [Assistant Professor Morgane Austern](/news/professor-morgane-austern-interviewed-international-day-women-statistics-and-data), [Professor Xihong Lin](/news/professor-xihong-lins-interview-international-day-women-statistics-and-data), and [Associate Professor Tracy Ke](/news/professor-tracy-ke-interviewed-international-day-women-statistics-and-data-science), illuminated their important work on statistical methods in a wide range of areas. With this series, we hope that younger scholars, particularly young women scholars, will read these interviews and feel empowered to pursue coursework, research, and careers in statistics. Since our previous interviews focused on strategies for cultivating future generations of women statisticians, it is appropriate that our last interview in this series is with [Dr. Kelly W. Zhang,](/people/kelly-w-zhang) a 2023 Harvard Computer Science PhD graduate who is currently a Postdoctoral Fellow with Professor Susan Murphy. As a PhD student, Dr. Zhang was advised by Professor Susan Murphy and Associate Professor Lucas Janson and worked in Prof. Murphy’s [Statistical Reinforcement Learning Lab](http://people.seas.harvard.edu/~samurphy/lab/overview.html). This fall, she will start a postdoc position at Columbia Business School before joining the Mathematics Department and the initiative Imperial-X at the Imperial College London as a faculty member. In the following edited and excerpted interview, Dr. Zhang shares her thoughts on the importance of having strong mentors and her work on developing statistical inference methods for reinforcement learning algorithms used in mobile health applications. ### 1. Please describe a research project that you’ve found interesting to work on in the past year. **Zhang:** My projects have been primarily motivated by the mobile health applications in Susan’s lab, where we design and deploy reinforcement learning algorithms for mobile health clinical trials. Reinforcement learning algorithms are sequential decision-making algorithms that update their decision-making process based on previously collected data. For instance, the algorithm might learn that sending users an activity suggestion to go for a walk when it’s cloudy outside is ineffective and will adjust future suggestions accordingly. In these mobile health clinical trials, the ability to perform various statistical analyses using the data collected by the reinforcement learning algorithm is critical. These analyses are essential for assessing the effectiveness of the digital health intervention and the reinforcement learning algorithm. My specific focus has been on developing statistical inference methods capable of handling data collected by reinforcement learning algorithms that learn from the combined information from multiple users. The advantage of such algorithms lies in their ability to reduce noise and expedite learning by leveraging data from multiple users rather than learning independently from each user. However, combining data across users introduces the challenge of dependence in the resulting user data. For instance, one user's responsiveness to an intervention, such as deciding to go for a walk based on a suggestion, will influence how the algorithm is updated, subsequently affecting suggestions made to other users. Consequently, standard statistical approaches that assume independent user data are not applicable in this context. I recently wrote a paper on a method to address this challenge with Lucas Janson and Susan Murphy, “[Statistical Inference After Adaptive Sampling for Longitudinal Data](https://arxiv.org/pdf/2202.07098.pdf).” This work is significant because it allows researchers to design clinical trials that use reinforcement algorithms that combine data across users for learning, while ensuring that researchers can perform rigorous statistical analyses on the resulting collected data. This level of assurance is crucial when seeking funding and working with collaborators in other fields, as it guarantees the validity of the statistical analysis in these mobile health clinical trials. ### 2. What inspired you to go into the field? What motivates you in your everyday work? **Zhang:** I discovered that I enjoyed research when I spent time working in a lab as an undergrad. Working in the lab was a great opportunity to see the day-to-day life of PhD students and to learn about their different research projects. This helped me to decide to pursue a PhD in Computer Science, with a focus on natural language processing. During my first semester of my PhD, I took a statistics class that I found really interesting because it provided a framework for developing a deeper understanding of data science methods. When designing models for natural language processing, I often had questions about why a model worked, or why it didn’t work and what went wrong. Statistics gave me a pathway for developing a more in-depth understanding of data science and machine learning, which led me to eventually join Professor Susan Murphy’s lab. An aspect of Susan’s lab that I appreciate is the close relationship between theory and applications; theory problems are directly motivated by the problems that the lab encounters in designing reinforcement learning algorithms for mobile health clinical trials. In my everyday work in Susan’s lab, I’m motivated to increase my understanding and make a positive impact on humanity (at least as much as I can), specifically related to disease prevention. Having female role models has also had a big impact on me by encouraging me and showing me what a research career might look like. On a day-to-day basis, I see Professor Murphy and Professor Doshi-Velez (Professor of Computer Science at Harvard SEAS) in their labs and observe how they balance their research and life outside of work. In one-on-one conversations with Susan, I have also received valuable mentoring. I remember early on in my PhD when I felt stuck and unsure about what was an important research question to work on and approached Susan. She shared with me the insight that whether you’re a PhD student, postdoc, assistant professor, or a professor like her, you are always struggling with this question and just need to sit down, make a thoughtful decision about a potential project, and move forward. Susan has also demonstrated to me that you can work on research projects that make a positive impact in the world, which is a great source of inspiration! ### 3. Moving forward in your work, what other areas might you be interested in? **Zhang:** In terms of the applications of bandit and reinforcement learning algorithms, I’m interested in pursuing a few paths. One is continuing to work on mobile health applications, which can apply to a wide range of areas, such as mental health or physical wellness. The other direction that I find interesting is how researchers are using these tools to help evaluate policy questions. For example, researchers at the World Bank have been conducting experiments in countries using bandit and reinforcement learning algorithms to optimize for the policy that works best. In their paper “[Adaptive Experiments for Policy Choice: Phone Calls for Home Reading in Kenya](https://elibrary.worldbank.org/doi/abs/10.1596/1813-9450-10098)” Bruno Esposito and Anja Sautmann describe how they used a bandit algorithm to optimize the automated calls sent to parents in Kenya to encourage parents to read with their children. It would be very cool to collaborate on similar projects that have the potential to impact education, health, or other policies implemented in developing countries. ### 4. What advice would you give to women who are interested in pursuing a PhD in statistics or computer science and a career in those fields? **Zhang:** The main advice that I have, not limited to women but applicable to everyone, is to seek research experience early on by joining a lab during college. It's very common for it to take some time to find a lab, so I encourage people getting started to keep in mind that initial attempts may not always succeed and to try to not take this rejection personally. The first time I tried joining a lab, I emailed back and forth with a postdoc for almost a year without a position in the lab materializing. Later on, though, after studying very hard in a class with the professor and having conversations about the research, I got the opportunity to work in a totally different lab. Based on my experience, I would advise undergrad students to be persistent (don’t get discouraged if things don’t initially work out!), put in the effort in their courses, and develop relationships with their professors. Taking these actions can open research opportunities. Once you have joined a lab, the work will be challenging and you might want to quit early on, but it’s important to spend enough time in a lab, like a semester or year, so that you can properly assess if a research area is a good fit for you. As you continue to do research and explore different labs, it’s helpful to think about what you find to be important – what motivates you and keeps you going. **Stats:** We appreciate Dr. Zhang sharing her tips with young scholars on how to explore work as a researcher. In addition, we’ve enjoyed learning about her journey towards academia, a journey that has most recently included developing novel statistical tools for evaluating mobile health apps. Our hope is that young scholars, particularly young women scholars, will read Dr. Zhang’s words and feel inspired to try a new stats class or approach a faculty member about a research project to see where it might lead.