fairness-aware utility design

This study addresses a critical gap in algorithmic fairness research, which has predominantly focused on trade-offs between performance and fairness in prediction space while overlooking the real-world utilities of multiple stakeholders and welfare distribution across groups. The authors propose a novel multi-stakeholder framework grounded in welfare economics and distributive justice, formalizing fairness as the social planner’s utility and employing posterior multi-objective optimization to identify optimal trade-offs between decision-maker utility and societal fairness. For the first time, they characterize the fairness–performance Pareto frontier in utility space under both deterministic and randomized policies, theoretically demonstrating that randomization can yield strictly superior trade-offs under certain conditions. Empirical results confirm that simple randomized mechanisms leverage outcome uncertainty to enhance fairness–performance balance, offering a more transparent and equitable design paradigm for algorithmic decision systems.

The R2 indicator for bi-objective optimization lacks strict Pareto compatibility—i.e., adding a dominated solution may not increase the indicator value. Method: This paper proposes an analytical variant of R2 based on a continuous uniform Tchebycheff utility function. We theoretically prove that this continuous R2 exhibits strict Pareto compatibility for bi-objective problems: adding any non-dominated solution strictly increases the indicator, and adding any dominated solution necessarily increases it. We further devise an exact O(N log N) algorithm, enabling the first efficient and theoretically compliant unary quality assessment. Results: Experiments show that the proposed indicator achieves evaluation performance comparable to hypervolume (HV), yet with significantly higher computational efficiency. It thus fills a critical gap in bi-objective set-quality indicators by simultaneously offering rigorous theoretical guarantees and practical scalability.

This work addresses the challenge of efficiently approximating the Pareto front in multi-objective optimization (MOO). We propose a novel set-based optimization framework grounded in the R2 utility function, which reformulates MOO as a single-objective optimization problem over solution sets. We theoretically establish that the R2 utility is both monotonic and submodular—properties enabling a greedy algorithm with a guaranteed (1−1/e) approximation ratio. Integrating this with Bayesian optimization, our approach achieves efficient, high-coverage approximation of the Pareto front. Methodologically, it unifies scalarization, submodular optimization, and Bayesian optimization within a coherent framework. Empirical evaluation on multi-objective Bayesian optimization benchmarks demonstrates significant improvements in convergence speed and Pareto front quality, validating both its practical efficacy and theoretical advantages.

This paper studies the student-school assignment problem under capacity constraints and group-level fairness requirements, jointly optimizing individual utilities (e.g., preference rankings), school enrollment caps, and inter-group fairness—such as across ethnicity or geography—formulated either via concave objective functions or explicit group-wise constraints, and supporting arbitrary covering constraints to capture multi-criteria and ordinal optimization needs. We propose, for the first time, a unified algorithmic framework that integrates convex programming modeling with systematic rounding techniques, yielding tunable randomized or deterministic algorithms. These run in polynomial time and provide controlled trade-offs among utility loss, capacity violations, and fairness deviations. Theoretically, our approach achieves provable approximation guarantees and naturally generalizes to covering constraints and ranking-aware settings. It exhibits strong scalability and practical deployability.

Existing engineering optimization benchmarks are largely confined to low-dimensional, moderately multimodal problems, lacking realism and scalability. This paper introduces a single- and multi-objective optimization benchmark suite tailored to human-powered aircraft design, integrating high-fidelity aerodynamic and material mechanics models. It comprises 60 problems spanning varying difficulty levels and adjustable dimensions (2–50D), supporting both constrained and unconstrained settings. Key contributions include: (i) the first integration of moderate multimodality with high engineering fidelity; (ii) controllable complexity scaling via parametrized wing segmentation; and (iii) generation of diverse multi-objective instances featuring Pareto fronts of varied geometries (e.g., convex, concave, disconnected). Analytical constraint modeling and penalty methods yield equivalent unconstrained formulations. Numerical experiments confirm strong multimodality, computational efficiency, and representative Pareto front coverage—establishing a more realistic and challenging testbed for algorithm evaluation.

This study investigates the systematic trade-offs between model performance and group fairness in algorithmic decision-making. Framing binary prediction as a multi-objective optimization problem that jointly maximizes decision-maker utility and fairness, the work reveals that Pareto-optimal solutions are characterized by group-specific threshold rules with both upper and lower bounds—extending beyond prior approaches that consider only lower-bound thresholds. The proposed framework accommodates arbitrary utility functions, data distributions, and generalized fairness metrics, unifying preprocessing, in-processing, and post-processing fairness interventions within a single theoretical structure. Theoretically, the Pareto frontier depends solely on population characteristics, the utility function, and the fairness measure, independent of any specific learning algorithm, thereby establishing a foundational basis for evaluating and comparing fair decision systems.

This study addresses the challenge of selecting a single solution from the Pareto front in multi-objective engineering optimization through scalarization, noting that different scalarizing functions exhibit significant differences in attainability, preference articulation, and coverage of non-supported solutions. The authors systematically compare four normalized scalarization approaches—weighted sum, achievement scalarizing function, desirability function, and fuzzy logic—on both convex and concave Pareto fronts, evaluating their performance via analytical control experiments in terms of attainable regions, selection density, sensitivity, and parameter interpretability. The findings reveal structural limitations of the weighted sum method on concave fronts, while the other three methods effectively cover non-supported regions. Notably, the desirability function introduces nonlinear preference mapping, and the fuzzy approach enables reference-dependent and non-separable modeling of engineering preferences, offering new insights for scalarization under complex preference structures.

This work addresses the challenge in multi-objective reinforcement learning (MORL) of simultaneously achieving Pareto optimality and fairness under unknown or dynamically changing user preferences, while also lacking a diverse set of fair policies. To this end, the authors propose a unified multi-policy learning framework that integrates generalized Gini welfare functions with non-stationary and stochastic policies. They theoretically show that under concave piecewise-linear welfare functions, the set of fair policies forms a convex coverage set. Building on this insight, they develop three novel algorithms: state-augmented non-stationary policy learning, stochastic policy optimization, and convex coverage set approximation. Experimental results across multiple benchmark environments demonstrate that the proposed approach significantly outperforms existing MORL baselines, successfully generating a diverse set of fair, Pareto-optimal policies that span a broad spectrum of user preferences.

AI deployment in high-stakes societal domains raises pressing fairness concerns, yet existing methods struggle to reconcile diverse fairness notions—individual vs. group, outcome vs. opportunity—with predictive accuracy. Method: We propose a human-centered, unified fairness framework that systematically integrates these two conceptual dichotomies into a single mathematical formalism, supporting eight computable metrics and explicitly modeling both marginal and intersectional fairness assumptions. The framework enables value-sensitive, multi-stakeholder negotiation over weighted fairness objectives, lowering barriers for non-expert practitioners. Contribution/Results: Evaluated across four real-world domains—income prediction, criminal justice, credit scoring, and healthcare—the framework quantifies trade-offs among fairness metrics, facilitates context-aware, value-aligned AI deployment decisions, and significantly enhances the interpretability and operationalizability of fairness practice.

This work addresses the challenge of simultaneously ensuring predictive effectiveness and fairness in AI/ML systems deployed in sensitive domains, where black-box models often fail to adequately handle both unidimensional and intersectional biases. The authors propose a multi-objective evolutionary search framework that integrates random forests with data perturbation techniques to jointly optimize five performance metrics and six fairness criteria—including intersectional fairness—for the first time enabling concurrent improvement of both unidimensional and intersectional fairness. The approach yields a customizable Pareto frontier of fairness-effectiveness trade-offs, allowing stakeholders to select solutions aligned with their priorities. Evaluated across 11 real-world scenarios, the method significantly outperforms 26 baseline approaches, maintaining high predictive accuracy while achieving state-of-the-art performance in mitigating intersectional bias.