# Unlabeled Debiasing in Downstream Tasks via Class-wise Low Variance Regularization

arXiv: https://arxiv.org/abs/2409.19541 · HTML: https://mlsanity.com/paper/2409.19541/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 19 of its 28 references that are in the map.
- First posted on arXiv 29 Sep 2024.

## Builds on (most cited first, 19 of 19)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Parameter-Efficient Transfer Learning for NLP](https://mlsanity.com/paper/1902.00751/) (2019)
- [A Survey on Bias and Fairness in Machine Learning](https://mlsanity.com/paper/1908.09635/) (2019)
- [AdapterFusion: Non-Destructive Task Composition for Transfer Learning](https://mlsanity.com/paper/2005.00247/) (2020)
- [AudioLDM: Text-to-Audio Generation with Latent Diffusion Models](https://mlsanity.com/paper/2301.12503/) (2023)
- [Null It Out: Guarding Protected Attributes by Iterative Nullspace Projection](https://mlsanity.com/paper/2004.07667/) (2020)
- [TinyStories: How Small Can Language Models Be and Still Speak Coherent English?](https://mlsanity.com/paper/2305.07759/) (2023)
- [Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Setting](https://mlsanity.com/paper/1901.09451/) (2019)
- [Large Scale Crowdsourcing and Characterization of Twitter Abusive Behavior](https://mlsanity.com/paper/1802.00393/) (2018)
- [Adversarial Removal of Demographic Attributes from Text Data](https://mlsanity.com/paper/1808.06640/) (2018)
- [FairFil: Contrastive Neural Debiasing Method for Pretrained Text Encoders](https://mlsanity.com/paper/2103.06413/) (2021)
- [Debiased Contrastive Learning of Unsupervised Sentence Representations](https://mlsanity.com/paper/2205.00656/) (2022)
- [Generating Data to Mitigate Spurious Correlations in Natural Language Inference Datasets](https://mlsanity.com/paper/2203.12942/) (2022)
- [A Novel Estimator of Mutual Information for Learning to Disentangle Textual Representations](https://mlsanity.com/paper/2105.02685/) (2021)
- [Societal Biases in Retrieved Contents: Measurement Framework and Adversarial Mitigation for BERT Rankers](https://mlsanity.com/paper/2104.13640/) (2021)
- [Parameter-efficient Modularised Bias Mitigation via AdapterFusion](https://mlsanity.com/paper/2302.06321/) (2023)
- [Modular and On-demand Bias Mitigation with Attribute-Removal Subnetworks](https://mlsanity.com/paper/2205.15171/) (2022)
- [Disentangling Representations of Text by Masking Transformers](https://mlsanity.com/paper/2104.07155/) (2021)
- [Effective Controllable Bias Mitigation for Classification and Retrieval using Gate Adapters](https://mlsanity.com/paper/2401.16457/) (2024)

## Built on by (most cited first, 3 of 3)

- [EMO-Debias: Benchmarking Gender Debiasing Techniques in Multi-Label Speech Emotion Recognition](https://mlsanity.com/paper/2506.04652/) (2025)
- [Investigating Gender Bias in LLM-Generated Stories via Psychological Stereotypes](https://mlsanity.com/paper/2508.03292/) (2025)
- [CO-VADA: A Confidence-Oriented Voice Augmentation Debiasing Approach for Fair Speech Emotion Recognition](https://mlsanity.com/paper/2506.06071/) (2025)

## Abstract (arXiv)

Language models frequently inherit societal biases from their training data. Numerous techniques have been proposed to mitigate these biases during both the pre-training and fine-tuning stages. However, fine-tuning a pre-trained debiased language model on a downstream task can reintroduce biases into the model. Additionally, existing debiasing methods for downstream tasks either (i) require labels of protected attributes (e.g., age, race, or political views) that are often not available or (ii) rely on indicators of bias, which restricts their applicability to gender debiasing since they rely on gender-specific words. To address this, we introduce a novel debiasing regularization technique based on the class-wise variance of embeddings. Crucially, our method does not require attribute labels and targets any attribute, thus addressing the shortcomings of existing debiasing methods. Our experiments on encoder language models and three datasets demonstrate that our method outperforms existing strong debiasing baselines that rely on target attribute labels while maintaining performance on the target task.