I am broadly interested in trustworthy ML, currently with a focus on explainability and causality.
In my view, the field of explainability faces two fundamental challenges: First, explainability in itself is not a well-defined goal but rather conflates different incompatible subgoals. To make progress, the different subgoals must be disentangled. Second, explainability methods themselves are subject to interpretation, and their meaning is often misunderstood.
Throughout my PhD, we demonstrated that a causal perspective helps tackle these challenges. First, since many goals in the context of explainability are inherently causal, and thus we need causal language to formalize them. Second, since many explainability methods perform some form of intervention, and thus causality helps to derive interpretation rules. We specifically focused on understanding model-agnostic methods and their application for scientific inference and recourse.
In my PostDoc, I plan to continue this line of research. Feel free to reach out if you would like to exchange or collaborate!
Publications
Please visit my Google Scholar profile for an up-to-date overview of my publications.