Read the original document by opening this link in a new tab.
Table of Contents
ABSTRACT
1 INTRODUCTION
2 METHODOLOGY
2.1 Distributional Intervention
2.1.1 Label Shift
2.1.2 Attribute Label Shift
2.2 Decoupling Aleatoric and Epistemic Uncertainty
2.3 Improving Fairness-Utility Trade-off
3 THEORETICAL GUARANTEE TO IMPROVE THE TRADE-OFF
Summary
The document discusses the proposal of a solution to balance fairness and utility in classification tasks by leveraging aleatoric uncertainty. It introduces a model to improve fairness when aleatoric uncertainty is high and utility elsewhere. The approach involves distributional intervention, decoupling of uncertainty types, and optimizing the fairness-utility trade-off. The theoretical guarantees for improving the trade-off are also presented in the document.