Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CausalPre: Scalable and Effective Data Pre-Processing for Causal Fairness

Ying Zheng, Yangfan Jiang, Kian-Lee Tan

Abstract

Causal fairness in databases is crucial to preventing biased and inaccurate outcomes in downstream tasks. While most prior work assumes a known causal model, recent efforts relax this assumption by enforcing additional constraints. However, these approaches often fail to capture broader attribute relationships that are critical to maintaining utility. This raises a fundamental question: Can we harness the benefits of causal reasoning to design efficient and effective fairness solutions without relying on strong assumptions about the underlying causal model? In this paper, we seek to answer this question by introducing CausalPre, a scalable and effective causality-guided data pre-processing framework that guarantees justifiable fairness, a strong causal notion of fairness. CausalPre extracts causally fair relationships by reformulating the originally complex and computationally infeasible extraction task into a tailored distribution estimation problem. To ensure scalability, CausalPre adopts a carefully crafted variant of low-dimensional marginal factorization to approximate the joint distribution, complemented by a heuristic algorithm that efficiently tackles the associated computational challenge. Extensive experiments on benchmark datasets demonstrate that CausalPre is both effective and scalable, challenging the conventional belief that achieving causal fairness requires trading off relationship coverage for relaxed model assumptions.

CausalPre: Scalable and Effective Data Pre-Processing for Causal Fairness

Abstract

Causal fairness in databases is crucial to preventing biased and inaccurate outcomes in downstream tasks. While most prior work assumes a known causal model, recent efforts relax this assumption by enforcing additional constraints. However, these approaches often fail to capture broader attribute relationships that are critical to maintaining utility. This raises a fundamental question: Can we harness the benefits of causal reasoning to design efficient and effective fairness solutions without relying on strong assumptions about the underlying causal model? In this paper, we seek to answer this question by introducing CausalPre, a scalable and effective causality-guided data pre-processing framework that guarantees justifiable fairness, a strong causal notion of fairness. CausalPre extracts causally fair relationships by reformulating the originally complex and computationally infeasible extraction task into a tailored distribution estimation problem. To ensure scalability, CausalPre adopts a carefully crafted variant of low-dimensional marginal factorization to approximate the joint distribution, complemented by a heuristic algorithm that efficiently tackles the associated computational challenge. Extensive experiments on benchmark datasets demonstrate that CausalPre is both effective and scalable, challenging the conventional belief that achieving causal fairness requires trading off relationship coverage for relaxed model assumptions.

Paper Structure

This paper contains 29 sections, 4 theorems, 21 equations, 13 figures, 1 table, 3 algorithms.

Table of Contents

  1. Introduction
  2. Motivation
  3. Our Contributions
  4. Preliminaries
  5. Causal DAGs
  6. Causal Fairness
  7. Information Measures
  8. Problem Statement
  9. $\texttt{CausalPre}$
  10. Naive Approach
  11. Refining the Fair DAG
  12. Estimating Attribute Distribution
  13. Accelerating Clique Selection
  14. The Complete Framework
  15. Utility-Fairness Trade-off
  16. ...and 14 more sections

Key Result

Theorem 1

Given a causal DAG $\mathcal{G}$ over attributes $\mathcal{V}$, if every directed path from any sensitive attribute in $\mathcal{S}$ to the outcome attribute $O$ contains at least one admissible attribute in $\mathcal{A}$, then the corresponding classifier $\mathcal{M}$ is justifiably fair.

Figures (13)

  • Figure 1: High-level comparison of fairness-aware pre-processing strategies on the biased manual labor hiring dataset. The top-half of the figure shows existing DAG-based or CI-based schemes, while the bottom-half illustrates $\texttt{CausalPre}$.
  • Figure 2: Refinement of the required structural guideline.
  • Figure 3: Illustration of the marginal-based decomposition.
  • Figure 4: End-to-end performance on real-world datasets. The gray shaded area indicates invalid regions.
  • Figure 5: Parameter sensitivity analysis.
  • ...and 8 more figures

Theorems & Definitions (10)

  • Definition 1: $\mathcal{K}$-fair salimi2019interventional
  • Definition 2: Justifiable Fairness salimi2019interventional
  • Theorem 1: salimi2019interventional
  • Corollary 1
  • Proposition 1
  • proof : Proof Sketch
  • Theorem 2
  • proof : Proof Sketch
  • proof
  • proof