# Fairness in Machine Learning: A Survey

**Authors:** Simon Caton, Christian Haas  
**arXiv:** [2010.04053](https://arxiv.org/abs/2010.04053)  
**Date:** 2020-10-04  
**Summary page:** https://www.summarizepaper.com/en/paper/2010.04053v1/

## Key points

- Increasing use of Machine Learning technologies in various contexts
- Need to ensure that these methods do not result in unexpected social biases or discrimination
- Three widely accepted frameworks for mitigating biases and increasing fairness: pre-processing, in-processing, and post-processing methods
- 11 method areas within each framework
- Expansion of discussion beyond fairness in binary classification tasks to include regression, recommender systems, unsupervised learning, and natural language processing
- Selection of currently available open-source libraries for implementing fair Machine Learning models
- Four dilemmas for fairness research highlighting open challenges in the field
- Valuable insights into different approaches for mitigating biases and promoting fairness in Machine Learning
- Useful resource for newcomers and experienced practitioners seeking to address social implications associated with Machine Learning technologies.

## Summary

The article "Fairness in Machine Learning: A Survey" by Simon Caton and Christian Haas addresses the increasing use of Machine Learning technologies in various contexts and the need to ensure that these methods do not result in unexpected social biases, such as gender or ethnic bias, or discrimination against people with disabilities. The authors acknowledge the significant literature on mitigating bias and promoting fairness but note that it can be complex for newcomers to navigate. To provide an overview of different approaches to mitigating biases and increasing fairness in Machine Learning, the article organizes these approaches into three widely accepted frameworks: pre-processing, in-processing, and post-processing methods. Within each framework, the authors further subcategorize the approaches into 11 method areas. While much of the existing literature focuses on fairness in binary classification tasks, this survey expands the discussion to include other areas such as regression, recommender systems, unsupervised learning, and natural language processing. Additionally, the authors provide a selection of currently available open-source libraries that can aid researchers and practitioners in implementing fair Machine Learning models. The article concludes by summarizing four dilemmas for fairness research that highlight open challenges in this field. By addressing these dilemmas, researchers can further advance efforts towards achieving fairness in Machine Learning applications. Overall, this comprehensive survey provides valuable insights into different schools of thought and approaches for mitigating biases and promoting fairness in Machine Learning. It serves as a useful resource for both newcomers and experienced practitioners seeking to understand and address social implications associated with Machine Learning technologies.