There are several central challenges that span all big data applications to the medical domain. We are always thinking about these issues in our projects: 

​​

• Limited data: Despite nearly all health records being digital now, there is a disconnect between modern big data approaches and the clinical research cycle. Big data approaches often require troves of annotated data, which is often not the case for areas of active clinical research, where the phenomenon itself is still being defined or simply rare, and large-scale annotation is not feasible. We explore methods for overcoming this limitation.

​

• Data heterogeneity: Many different types of data are collected and aggregated as part of the electronic health record. These data can span different modalities (e.g. imaging, text), biological domains, and temporal scale. We are interested in developing pipelines and techniques to ingest, align, and jointly represent these data using modern and traditional data fusion approaches. 

​

• Equity and Diversity: We are committed to ensuring our models and findings are as free from bias as possible and leverage the diversity in our data and statistical techniques to do so. We are also always on the lookout for advances that help address bias in computational modeling.

​

• Privacy: It goes without saying that patient privacy is paramount in our work. We are interested in exploring methods for generating insights from data across institutions without the need for sharing raw data.