Partition-and-Debias: Agnostic Biases Mitigation via A Mixture of Biases-Specific Experts

TL;DR

This paper proposes the Partition-and-Debias (PnD) method, which uses multiple bias-specific experts and a gating mechanism to mitigate unknown biases in image classification, addressing the complex real-world bias scenarios.

Contribution

It introduces a novel agnostic biases mitigation framework that implicitly divides bias space and effectively removes unknown biases in image classification.

Findings

Demonstrated effectiveness on public benchmarks

Outperformed existing bias mitigation methods

Validated on constructed datasets

Abstract

Bias mitigation in image classification has been widely researched, and existing methods have yielded notable results. However, most of these methods implicitly assume that a given image contains only one type of known or unknown bias, failing to consider the complexities of real-world biases. We introduce a more challenging scenario, agnostic biases mitigation, aiming at bias removal regardless of whether the type of bias or the number of types is unknown in the datasets. To address this difficult task, we present the Partition-and-Debias (PnD) method that uses a mixture of biases-specific experts to implicitly divide the bias space into multiple subspaces and a gating module to find a consensus among experts to achieve debiased classification. Experiments on both public and constructed benchmarks demonstrated the efficacy of the PnD. Code is available at: https://github.com/Jiaxuan-Li/PnD.