UniFair: A unified fair clustering approach based on separation and compactness
Researchers have introduced UniFair, a novel framework designed to enhance fairness in clustering algorithms. This approach simultaneously optimizes for separation fairness, ensuring protected groups are distant from decision boundaries, and social fairness, which minimizes disparities in clustering costs across groups. UniFair employs gradient-based optimization for both $k$-means and deep clustering objectives, demonstrating reduced group disparities with only a minor impact on overall clustering accuracy across various datasets. AI
IMPACT Introduces a method to mitigate bias in AI-driven decision-making processes.