Adaptive Sensitive Reweighting to Mitigate Bias in Fairness-aware Classification

By Emmanouil Krasanakis, Eleftherios Spyromitros Xioufis, Symeon Papadopoulos, Yiannis Kompatsiaris. 2018

179 citations recorded by the source corpus.

Machine learning bias and fairness have recently emerged as key issues due to the pervasive deployment of data-driven decision making in a variety of sectors and services. It has often been argued that unfair classifications can be attributed to bias in training data, but previous attempts to 'repair' training data have led to limited success. To circumvent shortcomings prevalent in data repairing approaches, such as those that weight training samples of the sensitive group (e.g. gender, race, financial status) based on their misclassification error, we present a process that iteratively adapts training sample weights with a theoretically grounded model. This model addresses different kinds of bias to better achieve fairness objectives, such as trade-offs between accuracy and disparate impact elimination or disparate mistreatment elimination. We show that, compared to previous fairness-aware approaches, our methodology achieves better or similar trades-offs between accuracy and unfairness mitigation on real-world and synthetic datasets.

Subfields: Artificial Intelligence, Economics and Econometrics, Safety Research, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Economics and Econometrics

In Safety Research

In AI Ethics, Fairness & Society

Citations in the corpus

A Survey on Data Collection for Machine Learning: A Big Data - AI Integration Perspective (2018, background): “In the model fairness literature [145], one approach to reducing unfairness is to fix the data.”
Classification with Fairness Constraints: A Meta-Algorithm with Provable Guarantees (2018, background): “, for statistical parity [42, 60] and equalized odds [59].”
Algorithm fairness in artificial intelligence for medicine and healthcare (2023, background): “Importance weighting reweights infrequent samples belonging to protected subgroups 67,68,70–72 .”
Identifying and Correcting Label Bias in Machine Learning (2019, methodology): “Moreover, these approaches typically do not perform as well as the state-of-art and have thus far come with few theoretical guarantees (Krasanakis et al., 2018).”
Artificial intelligence for digital and computational pathology (2023, background): “P a t c h i m p o r t a n c e”
AdaFair: Cumulative Fairness Adaptive Boosting (2019, methodology): “Finally, [23] assume that data does not contain robust and fair labels and proceed to estimate the true labels by re-adjusting iteratively the instance weights while minimizing equalized odds.”
Trustworthy AI: From Principles to Practices (2021, methodology): “Similar to pre-processing methods, reweighting [195] and adversarial learning [68] can be used for in-processing, with the potential of either making use of the model parameters or predictions that are not yet fully optimized to more directly debias the model.”
Fair Transfer Learning with Missing Protected Attributes (2019, methodology): “Given the popularity of weighting methods for the covariate shift problem and in works on fairness [1, 15, 17], we focus in this paper on weighting as a means to address covariate shift and fairness jointly.”
Fairness in Machine Learning: A Survey (2020, unknown): “Reference [181] also notes that some classifiers may be unaffected by the presence or specific nuances of some biases, and others may be negatively affected by altering the training data in an attempt to mitigate them.”
A survey on datasets for fairness‐aware machine learning (2021, background): “However, the protected attributes are connected together by an in-direct edge, which could be a reason for a similar accuracy of fairness-aware models of the related work (Hu et al., 2020) and (Krasanakis et al., 2018).”