Bias-inducing geometries: an exactly solvable data model with fairness implications

TMLR Paper3168 Authors

Abstract:

Machine learning (ML) may be oblivious to human bias but it is not immune to its perpetuation. Marginalisation and iniquitous group representation are often traceable in the very data used for training, and may be reflected or even enhanced by the learning models.
In the present work, we aim to clarify the role played by data geometry in the emergence of ML bias. We introduce an exactly solvable high-dimensional model of data imbalance, where parametric control over the many bias-inducing factors allows for an extensive exploration of the bias inheritance mechanism.
Through the tools of statistical physics, we analytically characterise the typical properties of learning models trained in this synthetic framework and obtain exact predictions for the observables that are commonly employed for fairness assessment.
Simplifying the nature of the problem to its minimal components, we can retrace and unpack typical unfairness behaviour observed on real-world datasets.
Finally, we focus on the effectiveness of bias mitigation strategies, first by considering a loss-reweighing scheme, that allows for an implicit minimisation of different unfairness metrics and a quantification of the incompatibilities between existing fairness criteria. Then, we propose a mitigation strategy based on a matched inference setting that entails the introduction of coupled learning models. Our theoretical analysis of this approach shows that the coupled strategy can strike superior fairness-accuracy trade-offs.

Submission Length: Regular submission (no more than 12 pages of main content)

Changes Since Last Submission:

We updated the manuscript following the comments of the reviewers.
Beyond the typos that the reviewers spotted, the main changes concern clarity:
* We split Fig.1 and Fig.2;
* We added a recap table in the supplementary for the notation;
* We add clarification in the text;
* We improved the discussion section on limitations and possible extensions.

Assigned Action Editor: ~Han_Bao2

Submission Number: 3168