High School and Undergraduate Student Volunteers as an Imperfect Solution to Machine Learning Geoscience Research Needs

Abstract

The machine learning revolution presents geoscientists with exciting new opportunities for research, but is constrained by the need for large, high quality training data sets. Simultaneously, undergraduate and graduate program admissions have become increasingly competitive, pressuring high school and undergraduate students to differentiate themselves through involvement in research at earlier stages. Aligning these two interests provides mutually beneficial opportunities for both geoscientists and early‐stage students. We describe our experiences working with 20 early‐stage students to build a large training data set digitized from satellite images of meltwater drainage patterns on ice sheets. The intent of this Perspective is to share our experience and lessons learned with other machine learning researchers who, like us, may have minimal experience mentoring young volunteer researchers but may seek such partnerships for the first time in response to their machine learning training data set needs. These partnerships enabled creation of a powerful new machine learning model that would have otherwise been infeasible. Student benefits varied with their commitment and proactiveness, ranging from exposure to geoscience research and a resume line item to strong letters of recommendation and ongoing connections with geoscience researchers at an elite university lab. Many students were attracted to the project solely out of interest in machine learning, so the opportunity reached students who would not otherwise have conducted research in geoscience. Still, without incentives for researchers to engage less‐privileged students, our experience suggests that mutually beneficial partnerships between researchers and early‐stage students may exacerbate issues of inequality and lack of diversity within the geosciences.