mlsanityData to 7 Oct 2026

When does Bias Transfer in Transfer Learning?

arXiv 2207.02842 ↗First posted 6 Jul 2022cs.LGcited by 23

Not on the map. Why?

Abstract (arXiv)

Using transfer learning to adapt a pre-trained "source model" to a downstream "target task" can dramatically increase performance with seemingly no downside. In this work, we demonstrate that there can exist a downside after all: bias transfer, or the tendency for biases of the source model to persist even after adapting the model to the target class. Through a combination of synthetic and natural experiments, we show that bias transfer both (a) arises in realistic settings (such as when pre-training on ImageNet or other standard datasets) and (b) can occur even when the target dataset is explicitly de-biased. As transfer-learned models are increasingly deployed in the real world, our work highlights the importance of understanding the limitations of pre-trained source models. Code is available at https://github.com/MadryLab/bias-transfer

Key facts. Cited by 23 papers in the map (data through 7 Oct 2026). Builds on 29 of its 56 references that are in the map. First posted on arXiv 6 Jul 2022.

Builds on

29 of 56 references.

References · 29

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  2. 2020Language Models are Few-Shot Learnerscited by 32,199
  3. 2014Microsoft COCO: Common Objects in Contextcited by 17,498
  4. 2015Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networkscited by 9,617
  5. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  6. 2014Deep Learning Face Attributes in the Wildcited by 3,679
  7. 2016DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFscited by 3,239
  8. 2013Rich feature hierarchies for accurate object detection and semantic segmentationcited by 2,160
  9. 2013Fine-Grained Visual Classification of Aircraftcited by 1,701
  10. 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,430
  11. 2017BadNets: Identifying Vulnerabilities in the Machine Learning Model Supply Chaincited by 1,052
  12. 2017ChestX-ray8: Hospital-scale Chest X-ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseasescited by 969
17 more
  1. 2017Revisiting Unreasonable Effectiveness of Data in Deep Learning Eracited by 916
  2. 2018Exploring the Limits of Weakly Supervised Pretrainingcited by 604
  3. 2018Do Better ImageNet Models Transfer Better?cited by 560
  4. 2016R-FCN: Object Detection via Region-based Fully Convolutional Networkscited by 414
  5. 2020Noise or Signal: The Role of Image Backgrounds in Object Recognitioncited by 342
  6. 2020Do Adversarially Robust ImageNet Models Transfer Better?cited by 315
  7. 2017Certified Defenses for Data Poisoning Attackscited by 288
  8. 2019Label-Consistent Backdoor Attackscited by 285
  9. 2018Interactive Supercomputing on 40,000 Cores for Machine Learning and Data Analysiscited by 110
  10. 2020Don't Judge an Object by Its Context: Learning to Overcome Contextual Biascited by 91
  11. 2021A Comprehensive Study on Face Recognition Biases Beyond Demographicscited by 86
  12. 2021CheXtransfer: Performance and Parameter Efficiency of ImageNet Models for Chest X-Ray Interpretationcited by 57
  13. 2016Object Recognition with and without Objectscited by 39
  14. 2015Transfer Learning from Deep Features for Remote Sensing and Poverty Mappingcited by 37
  15. 20213DB: A Framework for Debugging Computer Vision Modelscited by 31
  16. 2020Adversarially-Trained Deep Nets Transfer Better: Illustration on Image Classificationcited by 18
  17. 2019Self-Driving Car Steering Angle Prediction Based on Image Recognitioncited by 14

Built on by

Cited by year

Citing papers per year: 2022: 2; 2023: 10; 2024: 7; 2025: 4.

23 papers.

Sort:

By area:

  1. 2022Fake it till you make it: Learning transferable representations from synthetic ImageNet clonescited by 171
  2. 2023Overwriting Pretrained Bias with Finetuning Datacited by 32
  3. 2022ModelDiff: A Framework for Comparing Learning Algorithmscited by 30
  4. 2023Understanding the Detrimental Class-level Effects of Data Augmentationcited by 12
  5. 2024BiasPruner: Debiased Continual Learning for Medical Image Classificationcited by 12
  6. 2023Bias in Pruned Vision Models: In-Depth Analysis and Countermeasurescited by 11
  7. 2023Bridging the Digital Divide: Performance Variation across Socio-Economic Factors in Vision-Language Modelscited by 11
  8. 2023Meta-learning in healthcare: A surveycited by 9
  9. 2024Ask Your Distribution Shift if Pre-Training is Right for Youcited by 7
  10. 2025AI Supply Chains: An Emerging Ecosystem of AI Actors, Products, and Servicescited by 7
  11. 2023Astroformer: More Data Might not be all you need for Classificationcited by 6
  12. 2024Annotations on a Budget: Leveraging Geo-Data Similarity to Balance Model Performance and Annotation Costcited by 5
11 more
  1. 2024No Detail Left Behind: Revisiting Self-Retrieval for Fine-Grained Image Captioningcited by 5
  2. 2025ALBAR: Adversarial Learning approach to mitigate Biases in Action Recognitioncited by 5
  3. 2024Federated and Transfer Learning for Cancer Detection Based on Image Analysiscited by 3
  4. 2023Variation of Gender Biases in Visual Recognition Models Before and After Finetuningcited by 2
  5. 2023TorchQL: A Programming Framework for Integrity Constraints in Machine Learningcited by 2
  6. 2024Injecting Undetectable Backdoors in Obfuscated Neural Networks and Language Modelscited by 2
  7. 2024Views Can Be Deceiving: Improved SSL Through Feature Space Augmentationcited by 2
  8. 2023Continual Learning in the Presence of Spurious Correlationcited by 1
  9. 2023Pick your Poison: Undetectability versus Robustness in Data Poisoning Attackscited by 1
  10. 2025Assessing Intersectional Bias in Representations of Pre-Trained Image Recognition Modelscited by 0
  11. 2025Diagnosing Shortcut-Induced Rigidity in Continual Learning: The Einstellung Rigidity Index (ERI)cited by 0