Authors: Xianli Zeng, Guang Cheng, Edgar Dobriban

Published on: February 05, 2024

Impact Score: 8.12

Arxiv code: Arxiv:2402.02817

Summary

• What is new: The paper introduces methods for Bayes-optimal fair classification that minimize classification error under fairness constraints, using new notions of linear and bilinear disparity measures.
• Why this is important: Machine learning algorithms can impact protected groups differently, creating fairness issues.
• What the research proposes: Developed methods for achieving fair classification by understanding and leveraging the properties of linear and bilinear disparity measures in conjunction with fairness constraints.
• Results: These methods are shown to control disparity directly while maintaining close to optimal fairness-accuracy trade-offs, performing better than existing algorithms.

Technical Details

Technological frameworks used: Bayes-optimal fair classification

Models used: Probabilistic classifiers, Neyman-Pearson lemma

Data used: Group-wise regression functions

Potential Impact

Tech companies using machine learning for decision-making processes, HR and recruitment software industries, loan and credit scoring sectors

Want to implement this idea in a business?

We have generated a startup concept here: EquiLearn.