mlsanityData to 7 Oct 2026

Entropy-based Attention Regularization Frees Unintended Bias Mitigation from Lists

arXiv 2203.09192 ↗First posted 17 Mar 2022cs.CLcited by 54

Not on the map. Why?

Abstract (arXiv)

Natural Language Processing (NLP) models risk overfitting to specific terms in the training data, thereby reducing their performance, fairness, and generalizability. E.g., neural hate speech detection models are strongly influenced by identity terms like gay, or women, resulting in false positives, severe unintended bias, and lower performance. Most mitigation techniques use lists of identity terms or samples from the target domain during training. However, this approach requires a-priori knowledge and introduces further bias if important terms are neglected. Instead, we propose a knowledge-free Entropy-based Attention Regularization (EAR) to discourage overfitting to training-specific terms. An additional objective function penalizes tokens with low self-attention entropy. We fine-tune BERT via EAR: the resulting model matches or exceeds state-of-the-art performance for hate speech classification and bias metrics on three benchmark corpora in English and Italian. EAR also reveals overfitting terms, i.e., terms most likely to induce bias, to help identify their effect on the model, task, and predictions.

Key facts. Cited by 54 papers in the map (data through 7 Oct 2026). Builds on 33 of its 60 references that are in the map. First posted on arXiv 17 Mar 2022.

Builds on

33 of 60 references.

References · 33

  1. 2017Attention Is All You Needcited by 59,932
  2. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  3. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  4. 2014Neural Machine Translation by Jointly Learning to Align and Translatecited by 4,530
  5. 2016Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddingscited by 1,304
  6. 2019What Does BERT Look At? An Analysis of BERT's Attentioncited by 994
  7. 2016Semantics derived automatically from language corpora contain human-like biasescited by 894
  8. 2020CrowS-Pairs: A Challenge Dataset for Measuring Social Biases in Masked Language Modelscited by 703
  9. 2019Attention is not not Explanationcited by 629
  10. 2013Generating Sequences With Recurrent Neural Networkscited by 549
  11. 2019The Woman Worked as a Babysitter: On Biases in Language Generationcited by 540
  12. 2019Quantifying the Carbon Emissions of Machine Learningcited by 516
21 more
  1. 2020Null It Out: Guarding Protected Attributes by Iterative Nullspace Projectioncited by 442
  2. 2019Nuanced Metrics for Measuring Unintended Bias with Real Data for Text Classificationcited by 396
  3. 2019Is Attention Interpretable?cited by 387
  4. 2019Revealing the Dark Secrets of BERTcited by 300
  5. 2019Don't Take the Easy Way Out: Ensemble Based Methods for Avoiding Known Dataset Biasescited by 279
  6. 2017Word Embeddings Quantify 100 Years of Gender and Ethnic Stereotypescited by 251
  7. 2020Learning from the Worst: Dynamically Generated Datasets to Improve Online Hate Detectioncited by 217
  8. 2018Adversarial Removal of Demographic Attributes from Text Datacited by 166
  9. 2018Reducing Gender Bias in Abusive Language Detectioncited by 152
  10. 2019Multilingual and Multi-Aspect Hate Speech Analysiscited by 133
  11. 2019On Identifiability in Transformerscited by 115
  12. 2021Debiasing Pre-trained Contextualised Embeddingscited by 107
  13. 2021Challenges in Automated Debiasing for Toxic Language Detectioncited by 95
  14. 2020Contextualizing Hate Speech Classifiers with Post-hoc Explanationcited by 84
  15. 2020Unmasking Contextual Stereotypes: Measuring and Mitigating BERT's Gender Biascited by 82
  16. 2021Stereotype and Skew: Quantifying Gender Bias in Pre-trained and Fine-tuned Language Modelscited by 57
  17. 2019What's in a Name? Reducing Bias in Bios without Access to Protected Attributescited by 51
  18. 2020Demographics Should Not Be the Reason of Toxicity: Mitigating Discrimination in Text Classifications with Instance Weightingcited by 41
  19. 2017What does Attention in Neural Machine Translation Pay Attention to?cited by 30
  20. 2021Effective Attention Sheds Light On Interpretabilitycited by 11
  21. 2018Hierarchical CVAE for Fine-Grained Hate Speech Classificationcited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 2; 2023: 14; 2024: 17; 2025: 14; 2026: 7.

54 papers.

Sort:

By area:

  1. 2023Bias and Fairness in Large Language Models: A Surveycited by 626
  2. 2023XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Modelscited by 477
  3. 2024The Power of Noise: Redefining Retrieval for RAG Systemscited by 275
  4. 2023A Survey on Fairness in Large Language Modelscited by 91
  5. 2024Self-Debiasing Large Language Models: Zero-Shot Recognition and Reduction of Stereotypescited by 58
  6. 2023Training-free Diffusion Model Adaptation for Variable-Sized Text-to-Image Synthesiscited by 55
  7. 2023Sociodemographic Bias in Language Models: A Survey and Forward Pathcited by 27
  8. 2024Attention Entropy is a Key Factor: An Analysis of Parallel Context Encoding with Full-attention-based Pre-trained Language Modelscited by 26
  9. 2024Can LLMs Recognize Toxicity? A Structured Investigation Framework and Toxicity Metriccited by 23
  10. 2024Deconstructing The Ethics of Large Language Models from Long-standing Issues to New-emerging Dilemmas: A Surveycited by 23
  11. 2024Walking in Others' Shoes: How Perspective-Taking Guides Large Language Models in Reducing Toxicity and Biascited by 17
  12. 2024Does Liking Yellow Imply Driving a School Bus? Semantic Leakage in Language Modelscited by 14
38 more
  1. 2025Towards Transformer-Based Aligned Generation with Self-Coherence Guidancecited by 14
  2. 2023Bias A-head? Analyzing Bias in Transformer-Based Language Model Attention Headscited by 13
  3. 2023Should We Attend More or Less? Modulating Attention for Fairnesscited by 11
  4. 2023Tackling Bias in Pre-trained Language Models: Current Trends and Under-represented Societiescited by 11
  5. 2024Explaining Modern Gated-Linear RNNs via a Unified Implicit Attention Formulationcited by 11
  6. 2023Target-Agnostic Gender-Aware Contrastive Learning for Mitigating Bias in Multilingual Machine Translationcited by 10
  7. 2023Robust Natural Language Understanding with Residual Attention Debiasingcited by 10
  8. 2023Weigh Your Own Words: Improving Hate Speech Counter Narrative Generation via Attention Regularizationcited by 7
  9. 2024The Lou Dataset -- Exploring the Impact of Gender-Fair Language in German Text Classificationcited by 5
  10. 2024Feint and Attack: Attention-Based Strategies for Jailbreaking and Protecting LLMscited by 5
  11. 2026Self-Explaining Hate Speech Detection with Moral Rationalescited by 5
  12. 2024Why Don't Prompt-Based Fairness Metrics Correlate?cited by 4
  13. 2025LLM Bias Detection and Mitigation through the Lens of Desired Distributionscited by 4
  14. 2023Progressive Feature Self-reinforcement for Weakly Supervised Semantic Segmentationcited by 3
  15. 2024The Life Cycle of Large Language Models: A Review of Biases in Educationcited by 3
  16. 2023Learning Hierarchical Relational Representations through Relational Convolutionscited by 2
  17. 2024Detecting Bias in Large Language Models: Fine-tuned KcBERTcited by 2
  18. 2024Data Bias According to Bipol: Men are Naturally Right and It is the Role of Women to Follow Their Leadcited by 2
  19. 2024Towards Understanding Task-agnostic Debiasing Through the Lenses of Intrinsic Bias and Forgetfulnesscited by 2
  20. 2024Divided by discipline? A systematic literature review on the quantification of online sexism and misogyny using a semi-automated approachcited by 2
  21. 2025LFTF: Locating First and Then Fine-Tuning for Mitigating Gender Bias in Large Language Modelscited by 2
  22. 2025FALCON: Fine-grained Activation Manipulation by Contrastive Orthogonal Unalignment for Large Language Modelcited by 1
  23. 2025MirrorShield: Towards Universal Defense Against Jailbreaks via Entropy-Guided Mirror Craftingcited by 1
  24. 2025Theories of "Sexuality" in Natural Language Processing Bias Researchcited by 1
  25. 2025Learning from Sufficient Rationales: Analysing the Relationship Between Explanation Faithfulness and Token-level Regularisation Strategiescited by 1
  26. 2025On the Role of Hidden States of Modern Hopfield Network in Transformercited by 1
  27. 2026TensorLens: End-to-End Transformer Analysis via High-Order Attention Tensorscited by 1
  28. 2026CORAL: Correspondence Alignment for Improved Virtual Try-Oncited by 1
  29. 2022Domain Classification-based Source-specific Term Penalization for Domain Adaptation in Hate-speech Detectioncited by 0
  30. 2022Transferring Knowledge via Neighborhood-Aware Optimal Transport for Low-Resource Hate Speech Detectioncited by 0
  31. 2023Let the Models Respond: Interpreting Language Model Detoxification Through the Lens of Prompt Dependencecited by 0
  32. 2024NLPGuard: A Framework for Mitigating the Use of Protected Attributes by NLP Classifierscited by 0
  33. 2025Token-Level Entropy Reveals Demographic Disparities in Large Language Modelscited by 0
  34. 2025Hatevolution: What Static Benchmarks Don't Tell Uscited by 0
  35. 2025FairLangProc: A Python package for fairness in NLPcited by 0
  36. 2025Controllable Pareto Trade-off between Fairness and Accuracycited by 0
  37. 2025Bridging Fairness and Explainability: Can Input-Based Explanations Promote Fairness in Hate Speech Detection?cited by 0
  38. 2025Generalizing to Unseen Disaster Events: A Causal Viewcited by 0

50 most cited of 54 All 54 (JSON)