Parameter-efficient Modularised Bias Mitigation via AdapterFusion

TL;DR

This paper introduces DAM, a modular, adapter-based bias mitigation method for large language models that enables on-demand debiasing without altering the core model, ensuring parameter efficiency and task performance.

Contribution

The paper presents DAM, a novel adapter-based approach for flexible, on-demand bias mitigation in language models, avoiding irreversible model updates.

Findings

DAM effectively reduces bias across multiple attributes.

It maintains task performance while preventing catastrophic forgetting.

The method is parameter-efficient and easily switchable.

Abstract

Large pre-trained language models contain societal biases and carry along these biases to downstream tasks. Current in-processing bias mitigation approaches (like adversarial training) impose debiasing by updating a model's parameters, effectively transferring the model to a new, irreversible debiased state. In this work, we propose a novel approach to develop stand-alone debiasing functionalities separate from the model, which can be integrated into the model on-demand, while keeping the core model untouched. Drawing from the concept of AdapterFusion in multi-task learning, we introduce DAM (Debiasing with Adapter Modules) - a debiasing approach to first encapsulate arbitrary bias mitigation functionalities into separate adapters, and then add them to the model on-demand in order to deliver fairness qualities. We conduct a large set of experiments on three classification tasks with gender, race, and age as protected attributes. Our results show that DAM improves or maintains the effectiveness of bias mitigation, avoids catastrophic forgetting in a multi-attribute scenario, and maintains on-par task performance, while granting parameter-efficiency and easy switching between the original and debiased models.