[Paper Review] Towards Fairness-Aware Multi-Objective Optimization

Guo Yu, Lianbo Ma+3 more|arXiv (Cornell University)|Jul 22, 2022

Health Systems, Economic Evaluations, Quality of Life4 citations

TL;DR

This paper introduces fairness-aware multi-objective optimization (FMO) by integrating user preferences into multi-objective optimization to balance fairness and utility in machine learning. It proposes a framework that models fairness as a constrained objective alongside performance metrics, demonstrating through case studies that fairness can be achieved without significant trade-offs in primary objectives like profit or accuracy.

ABSTRACT

Recent years have seen the rapid development of fairness-aware machine learning in mitigating unfairness or discrimination in decision-making in a wide range of applications. However, much less attention has been paid to the fairness-aware multi-objective optimization, which is indeed commonly seen in real life, such as fair resource allocation problems and data driven multi-objective optimization problems. This paper aims to illuminate and broaden our understanding of multi-objective optimization from the perspective of fairness. To this end, we start with a discussion of user preferences in multi-objective optimization and then explore its relationship to fairness in machine learning and multi-objective optimization. Following the above discussions, representative cases of fairness-aware multiobjective optimization are presented, further elaborating the importance of fairness in traditional multi-objective optimization, data-driven optimization and federated optimization. Finally, challenges and opportunities in fairness-aware multi-objective optimization are addressed. We hope that this article makes a small step forward towards understanding fairness in the context of optimization and promote research interest in fairness-aware multi-objective optimization.

Motivation & Objective

• To address the gap in fairness-aware optimization within multi-objective settings, especially in real-world applications like resource allocation and data-driven optimization.
• To bridge fairness-aware machine learning (FML) and preference-driven multi-objective optimization (PMO) by identifying shared principles in user-centric decision-making.
• To demonstrate that fairness can be integrated into multi-objective optimization without severely compromising primary objectives such as profit or accuracy.
• To highlight challenges and opportunities in fairness-aware multi-objective optimization across traditional, data-driven, and federated optimization settings.
• To advocate for a human-in-the-loop approach that incorporates stakeholder preferences to guide the search for fair and effective solutions on the Pareto front.

Proposed method

• Formalizes fairness as a multi-objective optimization problem where fairness metrics and utility (e.g., accuracy, profit) are treated as competing objectives.
• Adapts preference-based multi-objective optimization (PMO) techniques, using reference vectors or weight vectors to represent decision-makers’ relative priorities between fairness and utility.
• Introduces a framework that maps fairness notions from FML (e.g., individual and group fairness) into optimization objectives, enabling trade-off analysis.
• Employs evolutionary and scalarization-based optimization methods to navigate the Pareto front and identify solutions that satisfy user preferences.
• Applies the framework to real-world scenarios such as dynamic electric vehicle charging pricing, where fairness in price distribution is balanced with profit maximization.
• Uses case studies to validate that fairness improvements do not necessarily degrade primary objectives, supporting the feasibility of joint optimization.

Experimental results

Research questions

• RQ1 How can fairness notions from fairness-aware machine learning be formally integrated into multi-objective optimization frameworks?
• RQ2 To what extent can fairness be achieved in multi-objective optimization without significantly compromising primary objectives like accuracy or profit?
• RQ3 What role do user preferences play in guiding the search for fair and effective solutions in multi-objective optimization?
• RQ4 How do conflicting fairness measures (e.g., individual vs. group fairness) interact in multi-objective optimization settings?
• RQ5 What are the key challenges and opportunities in extending fairness-aware optimization to data-driven and federated optimization scenarios?

Key findings

• Fairness-aware multi-objective optimization is feasible and effective, as demonstrated in dynamic pricing for electric vehicle charging, where fairness was improved without sacrificing expected profit.
• The integration of user preferences via weight vectors enables targeted search for optimal trade-offs between fairness and utility on the Pareto front.
• Case studies show that fairness constraints can be incorporated into optimization without a substantial loss in primary performance metrics, supporting the practical viability of FMO.
• There is a strong conceptual and methodological overlap between fairness-aware machine learning and preference-driven multi-objective optimization, suggesting shared design principles.
• The paper identifies that fairness-aware multi-objective optimization remains underexplored, especially in federated and data-driven settings, highlighting a significant research gap.
• The framework supports iterative, stakeholder-informed optimization, enabling dynamic adjustment of fairness-utility trade-offs based on real-world feedback and preferences.