Inference-Time Selective Debiasing to Enhance Fairness in Text Classification Models

2024-07-27

AI Summary

This work addresses the challenge of simultaneously achieving high predictive accuracy and fairness in text classification models without retraining. We propose a dynamic, inference-time selective debiasing method. Our approach comprises three key components: (1) the first adaptation of selective classification to fairness-aware prediction; (2) a KL-divergence-based bias-aware mechanism that more precisely identifies biased instances than conventional uncertainty estimation; and (3) lightweight post-hoc debiasing via Linear Encoder-Decoder Adversarial Correction (LEACE). Crucially, our method requires no architectural modifications or changes to the training pipeline. Evaluated on multiple benchmarks, it improves fairness metrics by 12–28% while incurring only marginal accuracy degradation. As a result, it significantly narrows the performance gap between post-processing debiasing methods and more costly training- or pre-processing-based approaches.

Technology Category

• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Ethics, Bias, and Fairness
• Computer Vision: Bias, Fairness & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Social Networks and Social Media: Fairness and bias in social network and social media analysis

Abstract

We propose selective debiasing -- an inference-time safety mechanism designed to enhance the overall model quality in terms of prediction performance and fairness, especially in scenarios where retraining the model is impractical. The method draws inspiration from selective classification, where at inference time, predictions with low quality, as indicated by their uncertainty scores, are discarded. In our approach, we identify the potentially biased model predictions and, instead of discarding them, we remove bias from these predictions using LEACE -- a post-processing debiasing method. To select problematic predictions, we propose a bias quantification approach based on KL divergence, which achieves better results than standard uncertainty quantification methods. Experiments on text classification datasets with encoder-based classification models demonstrate that selective debiasing helps to reduce the performance gap between post-processing methods and debiasing techniques from the at-training and pre-processing categories.

Problem

• Enhance fairness in text classification
• Selective debiasing at inference time
• Reduce bias using KL divergence

Innovation

• Inference-time selective debiasing
• LEACE post-processing debiasing
• KL divergence bias quantification