BMBENCH: Empirical Benchmarking of Algorithmic Fairness in Machine Learning Models

Kleyton da Costa1, 2, Cristian Munoz1, Bernando Modenesi3, Franklin Fernandez1,2, Adriano Koshiyama1,
1Holistic AI, 2Pontifical Catholic University of Rio de Janeiro, 2University of Utah,
Paper arXiv Code Leaderboard
BMBENCH Image

BMBENCH framework and pipeline process.

Abstract

The development and assessment of bias mitigation methods require rigorous benchmarks. This paper introduces BMBench, a comprehensive benchmarking framework to evaluate bias mitigation strategies across multitask machine learning predictions (binary classification, multiclass classification, regression, and clustering). Our benchmark leverages state-of-the-art and proposed datasets to improve fairness research, offering a broad spectrum of fairness metrics for a robust evaluation of bias mitigation methods. We provide an open-source repository to allow researchers to test and refine their bias mitigation approaches easily, promoting advancements in the creation of fair machine learning models.

Visual Effects

Using nerfies you can create fun visual effects. This Dolly zoom effect would be impossible without nerfies since it would require going through a wall.

BibTeX

@article{dacosta2025bmbench,
  author    = {da Costa, K., Munoz, C., Modenesi, B., Fernandez, F., Koshiyama, A.},
  title     = {BMBENCH: Empirical Benchmarking of Algorithmic Fairness in Machine Learning Models},
  journal   = {ICCV},
  year      = {2025},
}