Unbiased Model Prediction Without Using Protected Attribute Information

2026-03-31

AI Summary

This work addresses the challenge of demographic bias in deep learning models, which often exhibit disparate performance across subgroups defined by protected attributes such as gender or age. While existing debiasing methods typically require access to these protected attributes—limiting their practical applicability—this paper proposes NPAD, a novel debiasing algorithm that operates without them. NPAD innovatively leverages non-protected attributes to guide fairness optimization by constructing a Debiasing via Attribute Cluster Loss (DACL) through attribute clustering and further incorporates a Filter Redundancy Loss (FRL) to reduce feature redundancy. Experimental results on the LFWA and CelebA datasets demonstrate that NPAD significantly reduces prediction disparities across gender and age subgroups, marking the first effective approach to achieve fairness without relying on protected attributes.

Abstract

The problem of bias persists in the deep learning community as models continue to provide disparate performance across different demographic subgroups. Therefore, several algorithms have been proposed to improve the fairness of deep models. However, a majority of these algorithms utilize the protected attribute information for bias mitigation, which severely limits their application in real-world scenarios. To address this concern, we have proposed a novel algorithm, termed as \textbf{Non-Protected Attribute-based Debiasing (NPAD)} algorithm for bias mitigation, that does not require the protected attribute information. The proposed NPAD algorithm utilizes the auxiliary information provided by the non-protected attributes to optimize the model for bias mitigation. Further, two different loss functions, \textbf{Debiasing via Attribute Cluster Loss (DACL)} and \textbf{Filter Redundancy Loss (FRL)} have been proposed to optimize the model for fairness goals. Multiple experiments are performed on the LFWA and CelebA datasets for facial attribute prediction, and a significant reduction in bias across different gender and age subgroups is observed.

Problem

• bias mitigation
• protected attributes
• fairness
• deep learning
• unbiased prediction

Innovation

• fairness
• debiasing
• protected attributes
• loss function
• deep learning