Transportable Representations for Out-of-distribution Generalization
Keywords: Causality, domain generalization
TL;DR: We study properties of a family of representations that are characterized based on theory of transportability.
Abstract: Building on the theory of causal transportability (Bareinboim & Pearl), we define in this paper the notion of ``transportable representations," and show that the out-of-distribution generalization risk of classifiers defined based on these representations can be bounded, considering that graphical assumptions about the underlying system are provided.
Submission Number: 81
Loading