MACHINE LEARNING FOR CAUSALITY
KONRAD KORDING – UNIVERSITY OF PENNSYLVANIA
Machine learning traditionally does not get at causality and causality research traditionally treats machine learning as a dangerous set of highly biased estimators. In my talk I will talk about our lab’s efforts to use machine learning as a component of more traditional quasi-experimental techniques. I will also discuss meta-learning approaches to causal inference, approaches where the estimators themselves are learned. I will lament the relative lack of interactions between the various subfields of the causal inference space.
Related Paper: https://openreview.net/forum?id=fW60Hu9yDve