Disentangled Representation with Causal Constraints for Counterfactual Fairness

TL;DR

This paper introduces CF-VAE, a novel method for learning structured latent representations with causal constraints to improve counterfactual fairness in predictive models, outperforming existing fairness techniques.

Contribution

The paper proposes a new variational autoencoder that incorporates causal constraints to learn disentangled, structured representations for counterfactual fairness.

Findings

CF-VAE achieves better fairness and accuracy than benchmark methods.

Structured representations enable downstream models to attain counterfactual fairness.

Theoretical demonstration of the importance of causal relationships in fair representation learning.

Abstract

Much research has been devoted to the problem of learning fair representations; however, they do not explicitly the relationship between latent representations. In many real-world applications, there may be causal relationships between latent representations. Furthermore, most fair representation learning methods focus on group-level fairness and are based on correlations, ignoring the causal relationships underlying the data. In this work, we theoretically demonstrate that using the structured representations enable downstream predictive models to achieve counterfactual fairness, and then we propose the Counterfactual Fairness Variational AutoEncoder (CF-VAE) to obtain structured representations with respect to domain knowledge. The experimental results show that the proposed method achieves better fairness and accuracy performance than the benchmark fairness methods.