MQ-OFL: Multi-sensitive Queue-based Online Fair Learning

Abstract

Recently, there has been growing interest in fairness considerations in Artificial Intelligence (AI) and AI-based systems, as the decisions made by AI applications may negatively impact individuals and communities with ethical or legal consequences. Indeed, it is crucial to ensure that decisions based on AI-based systems do not reflect discriminatory behavior toward certain individuals or groups. The development of approaches to handle these concerns is an active area of research. However, most existing methods process the data in offline settings and are not directly suitable for online learning from evolving data streams. Further, these techniques fail to take the effects of data skew, or so-called class imbalance, on fairness-aware learning into account. In addition, recent fairness-aware online learning supervised learning approaches focus on one sensitive attribute only, which may lead to subgroup discrimination. In a fair classification, the equality of fairness metrics across multiple overlapping groups must be considered simultaneously. In this paper, we address the combined problem of fairness-aware online learning from imbalanced evolving streams, while considering multiple sensitive attributes. We introduce the Multi-Sensitive Queue-based Online Fair Learning (MQ-OFL) algorithm, an online fairness-aware approach, which maintains valid and fair models over evolving stream. MQ-OFL changes the training distribution in an online fashion based on both stream imbalance and discriminatory behavior of the model evaluated over the historical stream. We compare our MQ-OFL method with state-of-art studies on real-world data sets, and present comparative insights on the performance.