Adversarial Scrubbing of Demographic Information for Text Classification

TL;DR

This paper introduces Adversarial Scrubber (ADS), a framework that effectively removes demographic information from language model representations while preserving task performance, using adversarial learning and MDL probing for evaluation.

Contribution

The paper proposes a novel adversarial learning framework for debiasing language model representations, with theoretical convergence guarantees and enhanced evaluation methods.

Findings

ADS reduces demographic attribute information in representations

ADS maintains high performance on target tasks

Theoretical analysis confirms convergence under certain conditions

Abstract

Contextual representations learned by language models can often encode undesirable attributes, like demographic associations of the users, while being trained for an unrelated target task. We aim to scrub such undesirable attributes and learn fair representations while maintaining performance on the target task. In this paper, we present an adversarial learning framework "Adversarial Scrubber" (ADS), to debias contextual representations. We perform theoretical analysis to show that our framework converges without leaking demographic information under certain conditions. We extend previous evaluation techniques by evaluating debiasing performance using Minimum Description Length (MDL) probing. Experimental evaluations on 8 datasets show that ADS generates representations with minimal information about demographic attributes while being maximally informative about the target task.