The paper presents a new framework called Counterfactual Fairness-Aware Domain Generalization with Sequential Autoencoder (CDSAE). This framework aims to improve the performance of machine learning systems when dealing with out-of-distribution data across various domains. It focuses on separating environmental information and sensitive attributes from the embedded representation of classification features. This separation enhances model generalization across diverse domains and addresses unfair classification issues. The authors use principles of causal inference to tackle these issues and incorporate fairness regularization to use semantic information for classification purposes. The effectiveness of the approach is demonstrated through empirical validation on synthetic and real-world datasets, showing improved accuracy while preserving fairness.
Publication date: 25 Sep 2023
Project Page: Unavailable
Paper: https://arxiv.org/pdf/2309.13005
