- This event has passed.
Learning interpretable models from biomedical data-Department of Bioengineering Seminar Series
February 26 @ 12:00 pm - 1:00 pm
William La Cava, Ph.D. Postdoctoral Fellow, Institute for Biomedical Informatics, University of Pennsylvania, Philadelphia PA
Learning interpretable models from biomedical data
In the age of big data and big models, explainability has become the crux of scientific insight. This talk will focus on addressing explainable modeling using randomized search heuristics that explicitly optimize the trade-off between model conciseness and accuracy. These methods have been applied to different fields of engineering, including wind energy and non-linear dynamical systems. Here, I will highlight applications to the biomedical sciences, where growing access to electronic health records has created a large potential to improve human health via predictive modeling. A major short-term challenge with these data is the automatic extraction of useful features from large, messy data collections. I will show an example in which randomized search heuristics have been used to develop simple models for screening patients at risk of treatment-resistant hypertension. Finally, I will discuss two long-term challenges: 1) constraining these models to be fair, i.e. to perform equivalently on under-represented populations, and 2) automating more of the computational workflow of today’s data scientist, including tasks like data cleaning, model selection, feature engineering, and code writing.
William La Cava is a postdoctoral fellow in the Institute for Biomedical Informatics at the University of Pennsylvania. He received his PhD in Mechanical Engineering from the University of Massachusetts Amherst. His work focuses on methods for interpretable, fair, and automated machine learning for biomedical informatics. He is particularly interested in developing methods for learning predictive models from electronic health records.