New PRA framework uses inequality-aversion parameters for fairer resource allocation

A team of researchers from the University of British Columbia and National University of Singapore have proposed PRA (Parameterized Resource Allocation), a novel framework for fair resource allocation under diversity constraints. Traditional approaches enforce diversity as hard constraints, which can overly restrict the solution space and lead to suboptimal allocations. Inspired by risk-aversion in economics, PRA introduces controllable inequality-aversion parameters that softly regulate group-level diversity, allowing system designers to explicitly trade off between fairness and efficiency. The team also developed APRA, an adaptive variant that accommodates additional application-specific constraints like budget limits or compatibility rules. They prove that both frameworks achieve optimality under any fairness metric, regardless of the type of extra constraints, marking a significant theoretical advance.

Extensive experiments across three real-world applications—e-commerce recommendation, housing assignment, and university course allocation—demonstrate that PRA and APRA consistently outperform existing baselines in both effectiveness and robustness. For example, in housing allocation, PRA achieved higher overall satisfaction scores while maintaining tighter control over group diversity compared to hard-constraint baselines. The work addresses a critical gap in algorithmic fairness: how to make group-level diversity constraints flexible without sacrificing theoretical guarantees. As platforms increasingly need to balance equity with operational efficiency, this parameterized approach offers a practical, generalizable solution that can be deployed across domains ranging from hiring to public resource distribution.