Systematic Evaluation of Predictive Fairness
Not on the map. Why?
Abstract (arXiv)
Mitigating bias in training on biased datasets is an important open problem. Several techniques have been proposed, however the typical evaluation regime is very limited, considering very narrow data conditions. For instance, the effect of target class imbalance and stereotyping is under-studied. To address this gap, we examine the performance of various debiasing methods across multiple tasks, spanning binary classification (Twitter sentiment), multi-class classification (profession prediction), and regression (valence prediction). Through extensive experimentation, we find that data conditions have a strong influence on relative model performance, and that general conclusions cannot be drawn about method efficacy when evaluating only on standard datasets, as is current practice in fairness research.
Key facts. Cited by 3 papers in the map (data through 14 Oct 2026). Builds on 31 of its 48 references that are in the map. First posted on arXiv 17 Oct 2022.
Builds on
31 of 48 references.
References · 31
- 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
- 2014Adam: A Method for Stochastic Optimizationcited by 37,240
- 2016Equality of Opportunity in Supervised Learningcited by 2,007
- 2014Certifying and removing disparate impactcited by 715
- 2018Gender Bias in Coreference Resolution: Evaluation and Debiasing Methodscited by 659
- 2020Null It Out: Guarding Protected Attributes by Iterative Nullspace Projectioncited by 455
- 2016Fairness Beyond Disparate Treatment & Disparate Impact: Learning Classification without Disparate Mistreatmentcited by 369
- 2018Learning Adversarially Fair and Transferable Representationscited by 353
- 2019Bias in Bios: A Case Study of Semantic Representation Bias in a High-Stakes Settingcited by 349
- 2019Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Themcited by 268
- 2020Measuring and Reducing Gendered Correlations in Pre-trained Modelscited by 224
- 2016Demographic Dialectal Variation in Social Media: A Case Study of African-American Englishcited by 196
19 more
- 2018Adversarial Removal of Demographic Attributes from Text Datacited by 170
- 2021Harms of Gender Exclusivity and Challenges in Non-Binary Representation in Language Technologiescited by 128
- 2017Using millions of emoji occurrences to learn any-domain representations for detecting sentiment, emotion and sarcasmcited by 118
- 2022Linear Adversarial Concept Erasurecited by 101
- 2020FairBatch: Batch Selection for Model Fairnesscited by 99
- 2021An Empirical Survey of the Effectiveness of Debiasing Techniques for Pre-trained Language Modelscited by 97
- 2019Conditional Learning of Fair Representationscited by 78
- 2018Towards Robust and Privacy-preserving Text Representationscited by 75
- 2019Training individually fair ML models with Sensitive Subspace Robustnesscited by 73
- 2022FairLex: A Multilingual Benchmark for Evaluating Fairness in Legal Text Processingcited by 49
- 2021Diverse Adversaries for Mitigating Bias in Trainingcited by 48
- 2019Average Individual Fairness: Algorithms, Generalization and Experimentscited by 46
- 2021Your fairness may vary: Pretrained language model fairness in toxic text classificationcited by 39
- 2021Evaluating Debiasing Techniques for Intersectional Biasescited by 30
- 2022Optimising Equal Opportunity Fairness in Model Trainingcited by 22
- 2021Fairness-aware Class Imbalanced Learningcited by 19
- 2021An Empirical Comparison of Bias Reduction Methods on Real-World Problems in High-Stakes Policy Settingscited by 8
- 2022Towards Equal Opportunity Fairness through Adversarial Learningcited by 8
- 2022fairlib: A Unified Framework for Assessing and Improving Classification Fairnesscited by 4
Built on by
Citing papers per year: 2023: 1; 2024: 2.
3 papers.