Exploring Multi-Task Learning for Fairness in Machine Learning Regression
Resumo
Ensuring fairness in machine learning is a critical concern for high-stakes domains, yet most fairness-aware Multi-Task Learning (MTL) frameworks overlook regression problems in favor of classification. This work extends these techniques to regression, proposing a novel MTL framework that optimizes for equitable continuous outcomes across demographic subgroups. Our method dynamically reweights task-specific gradients during training to reduce disparities without compromising predictive accuracy. Evaluated on two real-world datasets, our approach, in the best scenarios, reduces subgroup disparity by up to 94.9% while also improving overall regression performance by up to 32.6%. These findings highlight the significant potential of fairness-aware MTL for creating more inclusive and responsible machine learning applications in sensitive domains.
Referências
Caruana, R. (1997). Multitask learning. Machine learning, 28:41–75.
Hardt, M., Price, E., and Srebro, N. (2016). Equality of opportunity in supervised learning. Advances in neural information processing systems, 29.
Li, C., Ding, S., Zou, N., Hu, X., Jiang, X., and Zhang, K. (2023). Multi-task learning with dynamic re-weighting to achieve fairness in healthcare predictive modeling. Journal of biomedical informatics, 143:104399.
Li, C.-T., Hsu, C., and Zhang, Y. (2022). Fairsr: Fairness-aware sequential recommendation through multi-task learning with preference graph embeddings. ACM Transactions on Intelligent Systems and Technology (TIST), 13(1):1–21.
Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., and Galstyan, A. (2021). A survey on bias and fairness in machine learning. ACM computing surveys (CSUR), 54(6):1–35.
Oneto, L., Doninini, M., Elders, A., and Pontil, M. (2019). Taking advantage of multitask learning for fair classification. In Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society, pages 227–237.
Raumanns, R., Schouten, G., Pluim, J. P., and Cheplygina, V. (2024). Dataset distribution impacts model fairness: Single vs. multi-task learning. In MICCAI Workshop on Fairness of AI in Medical Imaging, pages 14–23. Springer.
Roy, A. and Ntoutsi, E. (2022). Learning to teach fairness-aware deep multi-task learning. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pages 710–726. Springer.
Wang, Y., Wang, X., Beutel, A., Prost, F., Chen, J., and Chi, E. H. (2021). Understanding and improving fairness-accuracy trade-offs in multi-task learning. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, pages 1748–1757.
Zanna, K. and Sano, A. (2024). Enhancing fairness and performance in machine learning models: A multi-task learning approach with monte-carlo dropout and pareto optimality. arXiv preprint arXiv:2404.08230.
