BriarPatches: Pixel-Space Interventions for Inducing Demographic Parity

TL;DR

BriarPatch is a user-level pixel-space intervention that obscures sensitive attributes in pre-trained classifiers, promoting demographic parity and enabling fairer predictions without requiring model developer intervention.

Contribution

The paper introduces BriarPatch, a novel pixel-space intervention that allows users to promote demographic parity in classifiers without needing access to model internals.

Findings

BriarPatch effectively obscures sensitive attributes in representations.

The intervention promotes demographic parity in downstream predictions.

It offers a user-level alternative to model-centric fairness methods.

Abstract

We introduce the BriarPatch, a pixel-space intervention that obscures sensitive attributes from representations encoded in pre-trained classifiers. The patches encourage internal model representations not to encode sensitive information, which has the effect of pushing downstream predictors towards exhibiting demographic parity with respect to the sensitive information. The net result is that these BriarPatches provide an intervention mechanism available at user level, and complements prior research on fair representations that were previously only applicable by model developers and ML experts.