Unlock the Potential of Counterfactually-Augmented Data in Out-Of-Distribution Generalization

Caoyun Fan, Wenqing Chen, Jidong Tian, Yitian Li, Hao He, Yaohui Jin

Abstract: Highlights•Exclusion of non-edited causal features causes CAD inefficiency in OOD generalization.•This inefficiency is analyzed in feature space by Fisher’s linear discriminant.•Two constraints based on CAD structural properties help to extract causal features.•CAD’s potential for OOD generalization is unlocked.