Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Harmless Rawlsian Fairness Regardless of Demographic Prior

Xuanqian Wang, Jing Li, Ivor W. Tsang, Yew-Soon Ong

TL;DR

The experimental findings indicate that regression tasks, which are relatively unexplored from literature, can achieve significant fairness improvement through VFair regardless of any prior, whereas classification tasks usually do not because of their quantized utility measurements.

Abstract

Due to privacy and security concerns, recent advancements in group fairness advocate for model training regardless of demographic information. However, most methods still require prior knowledge of demographics. In this study, we explore the potential for achieving fairness without compromising its utility when no prior demographics are provided to the training set, namely \emph{harmless Rawlsian fairness}. We ascertain that such a fairness requirement with no prior demographic information essential promotes training losses to exhibit a Dirac delta distribution. To this end, we propose a simple but effective method named VFair to minimize the variance of training losses inside the optimal set of empirical losses. This problem is then optimized by a tailored dynamic update approach that operates in both loss and gradient dimensions, directing the model towards relatively fairer solutions while preserving its intact utility. Our experimental findings indicate that regression tasks, which are relatively unexplored from literature, can achieve significant fairness improvement through VFair regardless of any prior, whereas classification tasks usually do not because of their quantized utility measurements. The implementation of our method is publicly available at \url{https://github.com/wxqpxw/VFair}.

Towards Harmless Rawlsian Fairness Regardless of Demographic Prior

TL;DR

The experimental findings indicate that regression tasks, which are relatively unexplored from literature, can achieve significant fairness improvement through VFair regardless of any prior, whereas classification tasks usually do not because of their quantized utility measurements.

Abstract

Due to privacy and security concerns, recent advancements in group fairness advocate for model training regardless of demographic information. However, most methods still require prior knowledge of demographics. In this study, we explore the potential for achieving fairness without compromising its utility when no prior demographics are provided to the training set, namely \emph{harmless Rawlsian fairness}. We ascertain that such a fairness requirement with no prior demographic information essential promotes training losses to exhibit a Dirac delta distribution. To this end, we propose a simple but effective method named VFair to minimize the variance of training losses inside the optimal set of empirical losses. This problem is then optimized by a tailored dynamic update approach that operates in both loss and gradient dimensions, directing the model towards relatively fairer solutions while preserving its intact utility. Our experimental findings indicate that regression tasks, which are relatively unexplored from literature, can achieve significant fairness improvement through VFair regardless of any prior, whereas classification tasks usually do not because of their quantized utility measurements. The implementation of our method is publicly available at \url{https://github.com/wxqpxw/VFair}.

Paper Structure

This paper contains 28 sections, 3 theorems, 19 equations, 7 figures, 11 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related work
  3. VFair methodology
  4. Problem setup
  5. Fairness via minimizing variance of losses
  6. Objective formulation
  7. Harmless fairness update
  8. Experiments
  9. Experimental setup
  10. Examine harmless fairness in regression tasks
  11. Examine harmless fairness in classification tasks
  12. A closer look at VFair
  13. Conclusion
  14. Proof of Proposition 1
  15. Fairness objective
  16. ...and 13 more sections

Key Result

Proposition 1

$u \perp s$ holds for any $s$ that splits data into a number of groups, if and only if the loss $\ell$ is (approximately) independent of the training example $z$, i.e., $\ell \perp z$.

Figures (7)

  • Figure 1: Illustration of our idea through different forms of loss curves.
  • Figure 2: Two situations when updating primary and secondary gradient simultaneously.
  • Figure 3: Per-example losses for all compared methods sorted in ascending order on train set.
  • Figure 4: Experimental verification of the harmless update strategy.
  • Figure 5: Full version of per-example losses for all compared methods sorted in ascending order on the training set of four benchmark classification datasets. Dash lines represent their average losses.
  • ...and 2 more figures

Theorems & Definitions (4)

  • Proposition 1
  • Theorem 1
  • Theorem 2
  • proof