Online Learning with Multiple Fairness Regularizers via Graph-Structured Feedback
Researchers have developed a new online learning method designed to manage multiple, potentially conflicting, fairness objectives in automated decision systems. This approach is particularly useful when the optimal weighting of these fairness measures is unknown and needs to be learned adaptively over time through sequential interactions. The method operates within a bandit setting, utilizing graph-structured feedback to inform its adaptive learning process. AI
IMPACT Introduces a novel adaptive learning technique for managing complex fairness constraints in AI systems.