Why knowledge distillation amplifies gender bias and how to mitigate from the perspective of DistilBERT
Open accessAbstract
Knowledge distillation is widely used to transfer the language understanding of a large model to a smaller model.However, after knowledge distillation, it was found that the smaller model is more biased by gender compared to the source large model.This paper studies what causes gender bias to increase after the knowledge distillation process.Moreover, we suggest applying a variant of the mixup on knowledge distillation, which is used to increase generalizability during the distillation process, not for augmentation.By doing so, we can significantly reduce the gender bias amplification after knowledge distillation.We also conduct an experiment on the GLUE benchmark to demonstrate that even if the mixup is applied, it does not have a significant adverse effect on the model's performance.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
During knowledge distillation, the student model may inherit or even amplify biases from the teacher.
“During distillation, the student model may inherit or even amplify biases from the teacher (Ahn et al. 2022; Silva, Tambwekar, and Gombolay 2021).”
From Bias and Fairness in Large Language Models: A Survey (2024) · cites this paper as Ahn et al. 2022
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter
Sanh, V., Debut, L., Chaumond, J., & Wolf, T. (2019) · arXiv · doi:10.48550/arxiv.1910.01108 · Cited by 4,653 · see citing works
On Measuring Social Biases in Sentence Encoders
May, C., Wang, A., Bordia, S., Bowman, S. R., & Rudinger, R. (2019) · Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) · doi:10.18653/v1/n19-1063 · Cited by 461 · see citing works
Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting
De-Arteaga, M., Romanov, A., Wallach, H., Chayes, J., Borgs, C., Chouldechova, A., Geyik, S., Kenthapadi, K., & Kalai, A. T. (2019) · Proceedings of the Conference on Fairness, Accountability, and Transparency · doi:10.1145/3287560.3287572 · Cited by 337 · see citing works
Flagged first, then by citations
Research works printed but not linked: 15
- Su Lin Blodgett, Solon Barocas, Hal Daumé III, and Hanna Wallach. 2020. Language (technology) is power: A critical survey of" bias" in nlp. arXiv preprint arXiv:2005.14050. · arXiv
- Tolga Bolukbasi, Kai-Wei Chang, James Zou, Venkatesh Saligrama, and Adam Kalai. 2016. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. In Proceedings of the 30th International Conference on Neural Information Processing Systems, NIPS'16, page 4356-4364, Red Hook, NY, USA.
- Pengyu Cheng, Weituo Hao, Siyang Yuan, Shijing Si, and Lawrence Carin. 2021. Fairfil: Contrastive neu- ral debiasing method for pretrained text encoders. In International Conference on Learning Representa- tions.
- Ching-Yao Chuang and Youssef Mroueh. 2021. Fair mixup: Fairness via interpolation. arXiv preprint arXiv:2103.06503. · arXiv
- Mengnan Du, Subhabrata Mukherjee, Guanchu Wang, Ruixiang Tang, Ahmed Awadallah, and Xia Hu. 2021. Fairness via representation neutralization. Advances in Neural Information Processing Systems, 34.
- Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. 2015. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531. · arXiv
- Sara Hooker, Nyalleng Moorosi, Gregory Clark, Samy Bengio, and Emily Denton. 2020. Characteris- ing bias in compressed models. arXiv preprint arXiv:2010.03058. · arXiv
- Masahiro Kaneko and Danushka Bollegala. 2021. De- biasing pre-trained contextualised embeddings. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Lin- guistics: Main Volume, pages 1256-1266, Online. Association for Computational Linguistics. · DOI
- Kevin J Liang, Weituo Hao, Dinghan Shen, Yufan Zhou, Weizhu Chen, Changyou Chen, and Lawrence Carin. 2020a. Mixkd: Towards efficient distilla- tion of large-scale language models. arXiv preprint arXiv:2011.00593. · arXiv
- Paul Pu Liang, Irene Mengze Li, Emily Zheng, Yao Chong Lim, Ruslan Salakhutdinov, and Louis- Philippe Morency. 2020b. Towards debiasing sen- tence representations. In Proceedings of the 58th An- nual Meeting of the Association for Computational Linguistics, pages 5502-5515, Online. Association for Computational Linguistics. · DOI
- Luke McGuire, Tina Monzavi, Adam J. Hoffman, Fi- delia Law, Matthew J. Irvin, Mark Winterbottom, Adam Hartstone-Rose, Adam Rutland, Karen P. Burns, Laurence Butler, Marc Drews, Grace E. Fields, and Kelly Lynn Mulvey. 2021. Science and math in- terest and gender stereotypes: The role of educator gender in informal science learning sites. Frontiers in Psychology, 12. · DOI
- Victor Sanh, Thomas Wolf, Yonatan Belinkov, and Alexander M Rush. 2021. Learning from others' mistakes: Avoiding dataset biases without modeling them. In International Conference on Learning Rep- resentations.
- Andrew Silva, Pradyumna Tambwekar, and Matthew Gombolay. 2021. Towards a comprehensive under- standing and accurate evaluation of societal biases in pre-trained transformers. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2383-2389, Online. Association for Computational Linguistics. · DOI · Paper page
- Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman. 2019. GLUE: A multi-task benchmark and analysis plat- form for natural language understanding. In Interna- tional Conference on Learning Representations.
- Hongyi Zhang, Moustapha Cisse, Yann N. Dauphin, and David Lopez-Paz. 2018. mixup: Beyond empirical risk minimization. In International Conference on Learning Representations.
Cited by
Sorted by citations
Bias and Fairness in Large Language Models: A Survey
Gallegos, I. O., Rossi, R. A., Barrow, J., Tanjim, M. M., Kim, S., Dernoncourt, F., Yu, T., Zhang, R., & Ahmed, N. K. (2024) · Computational Linguistics · doi:10.1162/coli_a_00524 · Cited by 665 · see citing works
- Authors
- Ahn, J., Lee, H., Kim, J., & Oh, A.
- Affiliations
- Korea Advanced Institute of Science and Technology (KR)
- ORCID
- Venue
- Proceedings of the 4th Workshop on Gender Bias in Natural Language Processing (GeBNLP)
- Published
- 2022
- Pages
- 266-272
- OpenAlex
- W4287855127
- Cited by
- 18
- References
- 3
- Citations by year
- 2025: 4 · 2024: 7 · 2023: 6 · 2022: 1
- FWCI
- 4.11
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Workshop on Gender Bias in Natural Language Processing (GeBNLP) (publishedVersion, cc-by)
- Funders
- Korea Advanced Institute of Science and Technology
- Topics
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- knowledge distillation, gender bias, bias amplification, DistilBERT, mixup, bias mitigation, language model compression, GLUE benchmark
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
