Towards Equal Opportunity Fairness through Adversarial Learning
This work addresses fairness in NLP for users affected by biased models, but it is incremental as it builds on existing adversarial training approaches.
The paper tackled the problem of bias mitigation in natural language processing by proposing an augmented discriminator for adversarial training to explicitly model equal opportunity fairness, resulting in substantial improvements over standard methods in the performance-fairness trade-off across two datasets.
Adversarial training is a common approach for bias mitigation in natural language processing. Although most work on debiasing is motivated by equal opportunity, it is not explicitly captured in standard adversarial training. In this paper, we propose an augmented discriminator for adversarial training, which takes the target class as input to create richer features and more explicitly model equal opportunity. Experimental results over two datasets show that our method substantially improves over standard adversarial debiasing methods, in terms of the performance--fairness trade-off.