Test-Time Collective Action: Proxy-Based Perturbations for Correcting Algorithmic Harms

2026-05-26

AI Summary

This work addresses the challenge that users often lack effective means to mitigate algorithmic unfairness affecting specific subgroups without platform intervention. It proposes a test-time collective action framework wherein groups of users sharing query access collaboratively construct a surrogate model of the platform’s black-box API and generate transferable universal adversarial perturbations. These perturbations enable users to autonomously correct biased predictions at inference time, without requiring access to the original training pipeline. To the best of our knowledge, this is the first approach empowering users under black-box settings to collectively reduce performance disparities across subgroups through surrogate modeling and universal perturbation generation. Experiments on CIFAR-10, CIFAR-100, and FairFace demonstrate that the framework significantly improves worst-group accuracy, reduces equal opportunity gaps and disparate impact, and achieves these gains with lower collective query costs than individualized strategies.

Abstract

When machine learning systems under-perform for particular subgroups, affected users typically have no way to correct these disparities without relying on platform-level fixes. Existing approaches to algorithmic fairness rely on provider-centric approaches to correct these failures, leaving users with no external lever when faced with harm. Recent work in Algorithmic Collective Action shows that coordinated users can steer an algorithmic system toward a collective goal, but the existing mechanisms require the provider to retrain on the collective's modified data which users may not have control over. We propose Test-Time Collective Action (TTCA), a framework through which a group of users who share query access to the platform, can correct disparities affecting under-served subgroup without participating in the platform's training loop. We implement this through a proxy-based mechanism where the collective pools query access to a black-box API to extract a proxy of the platform, then optimizes a per-class universal perturbation against the proxy. Each member applies this perturbation to their own inputs at submission time, requiring no cooperation from the platform. We empirically evaluate the mechanism on CIFAR-10, CIFAR-100, and FairFace, showing that modestly-sized collectives close most of the subgroup accuracy gap, transfer across architectures (a small proxy can attack a larger platform), and improve worst-group accuracy, equal-opportunity gap, and disparate impact. A query-budget analysis comparing a per-user black-box attack baseline shows that pooling is cheaper than each subgroup member attacking alone. Test-time collective action thus offers corrective intervention to users when platform-side remediation is unavailable or delayed.

Problem

• algorithmic fairness
• subgroup disparities
• user-level intervention
• collective action
• test-time correction

Innovation

• Test-Time Collective Action
• Proxy-Based Perturbation
• Algorithmic Fairness
• Black-Box Attack
• Universal Perturbation