Post-processing for Fair Regression via Explainable SVD
This addresses fairness in regression models for applications where sensitive attributes should not be used during deployment, though it appears incremental as it builds on existing post-processing and SVD techniques.
The paper tackles the problem of training fair neural network regression models by developing a post-processing algorithm that uses explainable singular value decomposition to enforce statistical parity constraints, achieving similar or superior fairness-accuracy trade-offs compared to baselines without requiring sensitive attributes at inference time.
This paper presents a post-processing algorithm for training fair neural network regression models that satisfy statistical parity, utilizing an explainable singular value decomposition (SVD) of the weight matrix. We propose a linear transformation of the weight matrix, whereby the singular values derived from the SVD of the transformed matrix directly correspond to the differences in the first and second moments of the output distributions across two groups. Consequently, we can convert the fairness constraints into constraints on the singular values. We analytically solve the problem of finding the optimal weights under these constraints. Experimental validation on various datasets demonstrates that our method achieves a similar or superior fairness-accuracy trade-off compared to the baselines without using the sensitive attribute at the inference time.