mlsanityData to 7 Oct 2026

Overwriting Pretrained Bias with Finetuning Data

arXiv 2303.06167 ↗First posted 10 Mar 2023cs.CV cs.CY cs.LGcited by 32

Not on the map. Why?

Abstract (arXiv)

Transfer learning is beneficial by allowing the expressive features of models pretrained on large-scale datasets to be finetuned for the target task of smaller, more domain-specific datasets. However, there is a concern that these pretrained models may come with their own biases which would propagate into the finetuned model. In this work, we investigate bias when conceptualized as both spurious correlations between the target task and a sensitive attribute as well as underrepresentation of a particular group in the dataset. Under both notions of bias, we find that (1) models finetuned on top of pretrained models can indeed inherit their biases, but (2) this bias can be corrected for through relatively minor interventions to the finetuning dataset, and often with a negligible impact to performance. Our findings imply that careful curation of the finetuning dataset is important for reducing biases on a downstream task, and doing so can even compensate for bias in the pretrained model.

Key facts. Cited by 32 papers in the map (data through 7 Oct 2026). Builds on 54 of its 68 references that are in the map. First posted on arXiv 10 Mar 2023.

Builds on

54 of 68 references.

References · 54

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
  3. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  4. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 11,979
  5. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  6. 2019Momentum Contrast for Unsupervised Visual Representation Learningcited by 7,482
  7. 2014Deep Learning Face Attributes in the Wildcited by 3,679
  8. 2013Visualizing and Understanding Convolutional Networkscited by 2,278
  9. 2013Rich feature hierarchies for accurate object detection and semantic segmentationcited by 2,160
  10. 2016Equality of Opportunity in Supervised Learningcited by 2,000
  11. 2018Datasheets for Datasetscited by 1,371
  12. 2018Model Cards for Model Reportingcited by 1,068
38 more
  1. 2017BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chaincited by 1,052
  2. 2017A Closer Look at Memorization in Deep Networkscited by 936
  3. 2013DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognitioncited by 422
  4. 2017Men Also Like Shopping: Reducing Gender Bias Amplification using Corpus-level Constraintscited by 420
  5. 2022Last Layer Re-Training is Sufficient for Robustness to Spurious Correlationscited by 379
  6. 2018Rethinking ImageNet Pre-trainingcited by 376
  7. 2019Using Pre-Training Can Improve Model Robustness and Uncertaintycited by 373
  8. 2020The Pitfalls of Simplicity Bias in Neural Networkscited by 349
  9. 2021Multimodal datasets: misogyny, pornography, and malignant stereotypescited by 326
  10. 2020An Investigation of Why Overparameterization Exacerbates Spurious Correlationscited by 310
  11. 2020Data and its (dis)contents: A survey of dataset development and use in machine learning researchcited by 259
  12. 2020Large image datasets: A pyrrhic win for computer vision?cited by 237
  13. 2018Deep learning generalizes because the parameter-function map is biased towards simple functionscited by 180
  14. 2021Process for Adapting Language Models to Society (PALMS) with Values-Targeted Datasetscited by 179
  15. 2017No Classification without Representation: Assessing Geodiversity Issues in Open Data Sets for the Developing Worldcited by 152
  16. 2019Does Object Recognition Work for Everyone?cited by 149
  17. 2020Intrinsic Bias Metrics Do Not Correlate with Application Biascited by 148
  18. 2020REVISE: A Tool for Measuring and Mitigating Bias in Visual Datasetscited by 136
  19. 2020Learning from Failure: Training Debiased Classifier from Biased Classifiercited by 133
  20. 2020Fair Attribute Classification through Latent Space De-biasingcited by 125
  21. 2021Are Large-scale Datasets Necessary for Self-Supervised Pre-training?cited by 117
  22. 2019Lessons from Archives: Strategies for Collecting Sociocultural Data in Machine Learningcited by 116
  23. 2021Understanding and Evaluating Racial Biases in Image Captioningcited by 108
  24. 2020Image Representations Learned With Unsupervised Pre-Training Contain Human-like Biasescited by 107
  25. 2020Don't Judge an Object by Its Context: Learning to Overcome Contextual Biascited by 91
  26. 2022Picking on the Same Person: Does Algorithmic Monoculture lead to Outcome Homogenization?cited by 89
  27. 2019FairFace: Face Attribute Dataset for Balanced Race, Gender, and Agecited by 87
  28. 2021Do Datasets Have Politics? Disciplinary Values in Computer Vision Dataset Developmentcited by 85
  29. 2022A Whac-A-Mole Dilemma: Shortcuts Come in Multiples Where Mitigating One Amplifies Otherscited by 85
  30. 2020How Useful is Self-Supervised Pretraining for Visual Tasks?cited by 80
  31. 2022Vision Models Are More Robust And Fair When Pretrained On Uncurated Images Without Supervisioncited by 80
  32. 2022On the Intrinsic and Extrinsic Fairness Evaluation Metrics for Contextualized Language Representationscited by 75
  33. 2020Rethinking the Hyperparameters for Fine-tuningcited by 69
  34. 2020"What We Can't Measure, We Can't Understand": Challenges to Demographic Data Procurement in the Pursuit of Fairnesscited by 56
  35. 2021Directional Bias Amplificationcited by 54
  36. 2021Worst of Both Worlds: Biases Compound in Pre-trained Vision-and-Language Modelscited by 52
  37. 2021A Step Toward More Inclusive People Annotations for Fairnesscited by 46
  38. 2022Measuring Representational Harms in Image Captioningcited by 36

50 most cited of 54 All 54 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 2; 2024: 16; 2025: 10; 2026: 4.

32 papers.

Sort:

By area:

  1. 2024A Survey on Multilingual Large Language Models: Corpora, Alignment, and Biascited by 82
  2. 2023TIBET: Identifying and Evaluating Biases in Text-to-Image Generative Modelscited by 51
  3. 2024DeiT-LT Distillation Strikes Back for Vision Transformer Training on Long-Tailed Datasetscited by 14
  4. 2024Fairness and Bias Mitigation in Computer Vision: A Surveycited by 12
  5. 2023Inductive biases of multi-task learning and finetuning: multiple regimes of feature reusecited by 9
  6. 2024Mind the Graph When Balancing Data for Fairness or Robustnesscited by 9
  7. 2024Social Debiasing for Fair Multi-modal LLMscited by 8
  8. 2024Efficient Bias Mitigation Without Privileged Informationcited by 8
  9. 2024Large Language Models for Anomaly Detection in Computational Workflows: from Supervised Fine-Tuning to In-Context Learningcited by 6
  10. 2024Visual Data Diagnosis and Debiasing with Concept Graphscited by 6
  11. 2024SPD-CFL: Stepwise Parameter Dropout for Efficient Continual Federated Learningcited by 5
  12. 2024Using Self-supervised Learning Can Improve Model Fairnesscited by 5
20 more
  1. 2025Bias in Gender Bias Benchmarks: How Spurious Features Distort Evaluationcited by 5
  2. 2025The Impact of Coreset Selection on Spurious Correlations and Group Robustnesscited by 4
  3. 2024They're All Doctors: Synthesizing Diverse Counterfactuals to Mitigate Associative Biascited by 3
  4. 2024VideoLights: Feature Refinement and Cross-Task Alignment Transformer for Joint Video Highlight Detection and Moment Retrievalcited by 3
  5. 2025BiasConnect: Investigating Bias Interactions in Text-to-Image Modelscited by 2
  6. 2025From Global to Local: Social Bias Transfer in CLIPcited by 2
  7. 2025FairTune: A Bias-Aware Fine-Tuning Framework Towards Fair Heart Rate Prediction from PPGcited by 2
  8. 2024Common-Sense Bias Modeling for Classification Taskscited by 1
  9. 2026Noise-Aware Named Entity Recognition for Historical VET Documentscited by 1
  10. 2024Utilizing Adversarial Examples for Bias Mitigation and Accuracy Enhancementcited by 0
  11. 2024Towards Real-world Debiasing: Rethinking Evaluation, Challenge, and Solutioncited by 0
  12. 2024Dataset Representativeness and Downstream Task Fairnesscited by 0
  13. 2025A Critical Review of Predominant Bias in Neural Networkscited by 0
  14. 2025A Feature-level Bias Evaluation Framework for Facial Expression Recognition Modelscited by 0
  15. 2025Simulating Biases for Interpretable Fairness in Offline and Online Classifierscited by 0
  16. 2025Bi-CoG: Bi-Consistency-Guided Self-Training for Vision-Language Modelscited by 0
  17. 2025Representation Invariance and Allocation: When Subgroup Balance Matterscited by 0
  18. 2026When Agents Persuade: Rhetoric Generation and Mitigation in LLMscited by 0
  19. 2026The Global-Local loop: what is missing in bridging the gap between geospatial data from numerous communities?cited by 0
  20. 2026Two-Step Occupation Codingcited by 0