Fairness in Machine Learning: A Survey

Abstract

As Machine Learning technologies become increasingly used in contexts that affect citizens, companies as well as researchers need to be confident that their application of these methods will not have unexpected social implications, such as bias towards gender, ethnicity, and/or people with disabilities. There is significant literature on approaches to mitigate bias and promote fairness, yet the area is complex and hard to penetrate for newcomers to the domain. This article seeks to provide an overview of the different schools of thought and approaches to mitigating (social) biases and increase fairness in the Machine Learning literature. It organises approaches into the widely accepted framework of pre-processing, in-processing, and post-processing methods, subcategorizing into a further 11 method areas. Although much of the literature emphasizes binary classification, a discussion of fairness in regression, recommender systems, unsupervised learning, and natural language processing is also provided along with a selection of currently available open source libraries. The article concludes by summarising open challenges articulated as four dilemmas for fairness research.

Fair Structure Learning in Heterogeneous Graphical Models

“…(2018) and references therein. Our paper adds to the literature on fair methods for unsupervised learning tasks (Chierichetti et al, 2017;Celis et al, 2017;Samadi et al, 2018;Tantipongpipat et al, 2019;Oneto and Chiappa, 2020;Caton and Haas, 2020;Kleindessner et al, 2019). We discuss the work on fairness most closely related to our paper.…”

Learning Optimal Predictive Checklists

“…In this work, we presented a machine learning method to learn predictive checklists from data by solving an integer program. Our method illustrates a promising approach to fit models to obey constraints related to qualities like safety [1] and fairness [3,14]. Using our approach, practitioners can potentially co-design checklists alongside clinicians -by encoding requirements into the model and reporting their effects on predictive performance [17,36].…”

Deep Clustering based Fair Outlier Detection

“…Aside from fairness definitions, debiasing techniques can be categorized into three main approaches: pre-processing [13], in-processing [39,53] and post-processing [38]. The deep representation learning is a typical in-processing method that incorporates fairness constraints into the model optimization objectives to obtain an ideal model parameterization that maximizes performance and fairness [3]. In particular, representation learning with adversary has become a widely-used method in recent years and has demonstrated effectiveness on multiple tasks, including anonymization [11], clustering [32], classification [39], transfer learning [39], and domain adaptation [30,49].…”