Online Learning with Multiple Fairness Regularizers via Graph-Structured Feedback
A recent paper on machine learning tackles the difficulty of implementing various, frequently conflicting, fairness criteria in automated decision-making systems. The optimal balance of these fairness goals is usually not predetermined, can evolve over time, and needs to be learned adaptively through ongoing interactions. This research presents the issue as a bandit problem, utilizing graph-structured feedback for decision-making. The document can be found on arXiv within the computer science and machine learning sections.
Key facts
- Paper titled 'Online Learning with Multiple Fairness Regularizers via Graph-Structured Feedback'
- Addresses enforcing multiple competing fairness measures in automated decision systems
- Fairness objective weighting is unknown a priori and may change over time
- Weighting must be learned adaptively through sequential interactions
- Problem framed in a bandit setting with graph-structured feedback
- Available on arXiv with ID 2508.14311
- Submitted to Computer Science > Machine Learning category
- Includes references, citations, and code/data links
Entities
Institutions
- arXiv