Title: Partial Fairness Awareness: A Belief-Driven Strategic Approach for Strategic Agents

Original: arXiv:2606.00826v1 Announce Type: new Abstract: Strategic machine learning investigates scenarios where agents manipulate their features to receive favorable decisions from predictive models. To address fairness concerns intrinsic to strategic classification, recent work has introduced group-specific fairness constraints. However, current fairness-aware approaches face a fundamental dilemma in the issue of fairness exposure: making these constraints public enables strategic manipulation and can lead to fairness reversal, while keeping them hidden may reduce social welfare and discourage genuine improvement. To fill this gap, we subsequently propose the problem of partial fairness awareness (PFA), as our theoretical analysis informs that such a dilemma can be mitigated by releasing the candidate set of fairness constraints and concealing the grounding constraint. To be specific, we introduce a belief-guided strategic mechanism, wherein agents iteratively interact with the decision system and maintain a belief distribution over the candidate set of fairness constraints. This belief-guided process enables agents, through iterative interaction and feedback, to update their belief distribution over the candidate set, thereby gradually aligning their belief with the grounding fairness constraint employed by the system. Extensive experiments on real-world and synthetic datasets demonstrate that PFA achieves lower group fairness gaps, higher acceptance of truly qualified individuals, and more stable outcomes compared to fully public or private fairness regimes.

Rewrite: In strategic machine learning, agents often alter their attributes to secure advantageous outcomes from predictive algorithms. While recent studies have attempted to mitigate fairness issues in strategic classification by implementing group-specific constraints, existing fairness-aware methods encounter a critical challenge regarding transparency. Full disclosure of these constraints invites strategic gaming and potential fairness reversal, whereas complete opacity can diminish social welfare and stifle authentic self-improvement. To resolve this tension, we introduce the concept of Partial Fairness Awareness (PFA). Our theoretical findings suggest that this dilemma can be alleviated by publishing a set of candidate fairness constraints while keeping the actual, underlying constraint confidential. Specifically, we propose a belief-guided strategic mechanism in which agents engage in iterative exchanges with the decision-making system. During these interactions, agents cultivate a belief distribution across the published candidate constraints. Through continuous feedback and interaction, this process allows agents to refine their beliefs, ultimately converging toward the true fairness constraint utilized by the system. Our comprehensive evaluations on both real-world and synthetic data indicate that PFA outperforms fully transparent or fully opaque fairness models by delivering reduced group fairness disparities, improved acceptance rates for genuinely qualified candidates, and greater overall stability.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC