Exploring Fairness and Explainability in Multimodal Machine Learning Models Fusing Fitbit and Health Record Data Streams
Peter Washington, Professor
UC San Francisco
Applications for Spring 2025 are closed for this project.
This project involves building diagnostic machine learning models merging electronic health record (EHR) data with longitudinal Fitbit wearable data. After simply training these supervised learning models, the project will then focus on exploring explainable AI + algorithmic fairness issues with the dataset, all using Python-based machine learning. This project involves using an existing dataset (NIH's All of Us dataset). More information about the All of Us dataset can be found here: https://allofus.nih.gov/.
Role: The student will lead in the Python data exploration, cleaning, and analysis, followed by machine learning model development using Fitbit data. The ultimate goal is for the undergraduate student to lead a first-author publication in a leading venue (which will lead to a very competitive graduate school application); Dr. Washington will help the student achieve this goal.
Qualifications: A successful candidate will have at least 3 out of the 4 qualifications:
- Python (Numpy, Pandas, Matplotlib/Seaborn, etc)
- Machine learning (Scikitlearn, Tensorflow or Pytorch)
- Formal writing capability (writing an academic research paper)
- Motivation and enthusiasm for the intersection of machine learning + healthcare
Hours: to be negotiated
Off-Campus Research Site: Since this work is computational, most/all work can be conducted off campus. We can meet in person periodically for mentorship purposes.
Related website: https://peterwashington.net/
Related website: https://allofus.nih.gov/