Data, and how it's represented, is critical in developing performant and predictive models. My interest centers on leveraging inherently-compositional data representation frameworks to facilitate predictive models for biomedically relevant tasks (e.g. drug safety and cancer research). Similarly, I also investigate more traditional machine learning algorithms (e.g. SVMs, recurrentneural networks) in these problem contexts. As the biomedical literature is a rich, but complex repository of collected knowledge that's useful for causal inference - a particularly important consideration in the biomedical domain - I'm especially interested in using natural language data from the biomedical literature in developing these models.