Details
| Original language | English |
|---|---|
| Title of host publication | Discovery Science |
| Subtitle of host publication | 25th International Conference, DS 2022, Montpellier, France, October 10–12, 2022, Proceedings |
| Editors | Poncelet Pascal, Dino Ienco |
| Pages | 286-301 |
| Number of pages | 16 |
| ISBN (electronic) | 978-3-031-18840-4 |
| Publication status | Published - 2022 |
| Event | 25th International Conference on Discovery Science, DS 2022 - Montpellier, France Duration: 10 Oct 2022 → 12 Oct 2022 |
Publication series
| Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
|---|---|
| Volume | 13601 |
| ISSN (Print) | 0302-9743 |
| ISSN (electronic) | 1611-3349 |
Abstract
Recent studies showed that datasets used in fairness-aware machine learning for multiple protected attributes (referred to as multi-discrimination hereafter) are often imbalanced. The class-imbalance problem is more severe for the protected group in the critical minority class (e.g., female +, non-white +, etc.). Still, existing methods focus only on the overall error-discrimination trade-off, ignoring the imbalance problem, and thus they amplify the prevalent bias in the minority classes. To solve the combined problem of multi-discrimination and class-imbalance we introduce a new fairness measure, Multi-Max Mistreatment (MMM), which considers both (multi-attribute) protected group and class membership of instances to measure discrimination. To solve the combined problem, we propose Multi-Fair Boosting Post Pareto (MFBPP) a boosting approach that incorporates MMM-costs in the distribution update and post-training, selects the optimal trade-off among accurate, class-balanced, and fair solutions. The experimental results show the superiority of our approach against state-of-the-art methods in producing the best balanced performance across groups and classes and the best accuracy for the protected groups in the minority class.
Keywords
- Boosting, Class-imbalance, Multi-discrimination
ASJC Scopus subject areas
- Mathematics(all)
- Theoretical Computer Science
- Computer Science(all)
- General Computer Science
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
Discovery Science: 25th International Conference, DS 2022, Montpellier, France, October 10–12, 2022, Proceedings. ed. / Poncelet Pascal; Dino Ienco. 2022. p. 286-301 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13601).
Research output: Chapter in book/report/conference proceeding › Conference contribution › Research › peer review
}
TY - GEN
T1 - Multi-fairness Under Class-Imbalance
AU - Roy, Arjun
AU - Iosifidis, Vasileios
AU - Ntoutsi, Eirini
N1 - Funding Information: Acknowledgements. The work of the first author is supported by the Volkswagen Foundation under the call “Artificial Intelligence and the Society of the Future” (the BIAS project). We are sincerely thankful to the invaluable suggestion of Prof. Niloy Ganguly from L3S Research Center, in shaping up the paper to its current form. Most of the work was carried out while the last author was affiliated with Freie Universität Berlin, Germany.
PY - 2022
Y1 - 2022
N2 - Recent studies showed that datasets used in fairness-aware machine learning for multiple protected attributes (referred to as multi-discrimination hereafter) are often imbalanced. The class-imbalance problem is more severe for the protected group in the critical minority class (e.g., female +, non-white +, etc.). Still, existing methods focus only on the overall error-discrimination trade-off, ignoring the imbalance problem, and thus they amplify the prevalent bias in the minority classes. To solve the combined problem of multi-discrimination and class-imbalance we introduce a new fairness measure, Multi-Max Mistreatment (MMM), which considers both (multi-attribute) protected group and class membership of instances to measure discrimination. To solve the combined problem, we propose Multi-Fair Boosting Post Pareto (MFBPP) a boosting approach that incorporates MMM-costs in the distribution update and post-training, selects the optimal trade-off among accurate, class-balanced, and fair solutions. The experimental results show the superiority of our approach against state-of-the-art methods in producing the best balanced performance across groups and classes and the best accuracy for the protected groups in the minority class.
AB - Recent studies showed that datasets used in fairness-aware machine learning for multiple protected attributes (referred to as multi-discrimination hereafter) are often imbalanced. The class-imbalance problem is more severe for the protected group in the critical minority class (e.g., female +, non-white +, etc.). Still, existing methods focus only on the overall error-discrimination trade-off, ignoring the imbalance problem, and thus they amplify the prevalent bias in the minority classes. To solve the combined problem of multi-discrimination and class-imbalance we introduce a new fairness measure, Multi-Max Mistreatment (MMM), which considers both (multi-attribute) protected group and class membership of instances to measure discrimination. To solve the combined problem, we propose Multi-Fair Boosting Post Pareto (MFBPP) a boosting approach that incorporates MMM-costs in the distribution update and post-training, selects the optimal trade-off among accurate, class-balanced, and fair solutions. The experimental results show the superiority of our approach against state-of-the-art methods in producing the best balanced performance across groups and classes and the best accuracy for the protected groups in the minority class.
KW - Boosting
KW - Class-imbalance
KW - Multi-discrimination
UR - http://www.scopus.com/inward/record.url?scp=85142727007&partnerID=8YFLogxK
U2 - 10.48550/arXiv.2104.13312
DO - 10.48550/arXiv.2104.13312
M3 - Conference contribution
AN - SCOPUS:85142727007
SN - 9783031188398
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 286
EP - 301
BT - Discovery Science
A2 - Pascal, Poncelet
A2 - Ienco, Dino
T2 - 25th International Conference on Discovery Science, DS 2022
Y2 - 10 October 2022 through 12 October 2022
ER -