Speaker
Description
Machine learning (ML) models are increasingly deployed in high-stakes
environments, e.g., in the health domain, where ethical and legal
considerations require models to be interpretable. Despite substantial
progress in interpretable ML (IML), several key challenges remain. These
include distinguishing between interpreting the model and using the
model to interpret the data-generating process, dealing with
heterogeneous data, and providing meaningful recourse recommendations. I
argue that we can address these challenges by adopting a causal
perspective on IML. In particular, I demonstrate how the causal
structures of the model and the data-generating process jointly affect
model interpretation techniques. I then provide an overview of our
recent work on leveraging causal techniques to improve model
interpretation and render recourse recommendations more meaningful.