FairIF: Boosting Fairness in Deep Learning via Influence Functions with Validation Set Sensitive Attributes

Abstract

Most fair machine learning methods either highly rely on the sensitive information of the training samples or require a large modification on the target models, which hinders their practical application. To address this issue, we propose a two-stage training algorithm named FAIRIF. It minimizes the loss over the reweighted data set (second stage) where the sample weights are computed to balance the model performance across different demographic groups (first stage). FAIRIF can be applied on a wide range of models trained by stochastic gradient descent without changing the model, while only requiring group annotations on a small validation set to compute sample weights. Theoretically, we show that, in the classification setting, three notions of disparity among different groups can be mitigated by training with the weights. Experiments on synthetic data sets demonstrate that FAIRIF yields models with better fairness-utility trade-offs against various types of bias; and on real-world data sets, we show the effectiveness and scalability of FAIRIF. Moreover, as evidenced by the experiments with pretrained models, FAIRIF is able to alleviate the unfairness issue of pretrained models without hurting their performance.

References in corpus (12)

• Equality of Opportunity in Supervised Learning
• Fairness Beyond Disparate Treatment & Disparate Impact: Learning Classification without Disparate Mistreatment
• A Clarification of the Nuances in the Fairness Metrics Landscape
• Bias in Machine Learning Software: Why? How? What to do?
• Identifying and Correcting Label Bias in Machine Learning
• Studying Large Language Model Generalization with Influence Functions
• Dataset Pruning: Reducing Training Data by Examining Generalization Influence
• Fairness-aware Model-agnostic Positive and Unlabeled Learning
• Understanding Programmatic Weak Supervision via Source-aware Influence Function
• Achieving Fairness at No Utility Cost via Data Reweighing with Influence
• Fair Infinitesimal Jackknife: Mitigating the Influence of Biased Training Data Points Without Refitting
• Understanding Instance-Level Impact of Fairness Constraints