mlsanityData to 7 Oct 2026

Characterising Bias in Compressed Models

arXiv 2010.03058 ↗First posted 6 Oct 2020cs.LG cs.AIcited by 162

Not on the map. Why?

Abstract (arXiv)

The popularity and widespread use of pruning and quantization is driven by the severe resource constraints of deploying deep neural networks to environments with strict latency, memory and energy requirements. These techniques achieve high levels of compression with negligible impact on top-line metrics (top-1 and top-5 accuracy). However, overall accuracy hides disproportionately high errors on a small subset of examples; we call this subset Compression Identified Exemplars (CIE). We further establish that for CIE examples, compression amplifies existing algorithmic bias. Pruning disproportionately impacts performance on underrepresented features, which often coincides with considerations of fairness. Given that CIE is a relatively small subset but a great contributor of error in the model, we propose its use as a human-in-the-loop auditing tool to surface a tractable subset of the dataset for further inspection or annotation by a domain expert. We provide qualitative and quantitative support that CIE surfaces the most challenging examples in the data distribution for human-in-the-loop auditing.

Key facts. Cited by 162 papers in the map (data through 7 Oct 2026). First posted on arXiv 6 Oct 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 33; 2022: 31; 2023: 34; 2024: 32; 2025: 14; 2026: 18.

162 papers.

Sort:

By area:

  1. 2021AI and the Everything in the Whole Wide World Benchmarkcited by 340
  2. 2022Matryoshka Representation Learningcited by 268
  3. 2022Multimodal Contrastive Learning with LIMoE: the Language-Image Mixture of Expertscited by 202
  4. 2021Efficient Deep Learning: A Survey on Making Deep Learning Models Smaller, Faster, and Bettercited by 153
  5. 2021Deep Learning Through the Lens of Example Difficultycited by 139
  6. 2023Evaluating the Social Impact of Generative AI Systems in Systems and Societycited by 91
  7. 2023MatFormer: Nested Transformer for Elastic Inferencecited by 79
  8. 2022Efficient Methods for Natural Language Processing: A Surveycited by 78
  9. 2021Can Subnetwork Structure be the Key to Out-of-Distribution Generalization?cited by 75
  10. 2022The Effects of Regularization and Data Augmentation are Class Dependentcited by 67
  11. 2021BERT Learns to Teach: Knowledge Distillation with Meta Learningcited by 61
  12. 2021Randomness In Neural Network Training: Characterizing The Impact of Toolingcited by 57
38 more
  1. 2021How Well Do Sparse Imagenet Models Transfer?cited by 50
  2. 2021The Low-Resource Double Bind: An Empirical Study of Pruning for Low-Resource Machine Translationcited by 46
  3. 2023Pruning vs Quantization: Which is Better?cited by 45
  4. 2021Efficient Lottery Ticket Finding: Less Data is Morecited by 44
  5. 2022Pruning has a disparate impact on model accuracycited by 43
  6. 2022Membership Inference Attacks and Defenses in Neural Network Pruningcited by 42
  7. 2023Counting Carbon: A Survey of Factors Influencing the Emissions of Machine Learningcited by 42
  8. 2024Low-Rank Quantization-Aware Training for LLMscited by 42
  9. 2022The Combinatorial Brain Surgeon: Pruning Weights That Cancel One Another in Neural Networkscited by 41
  10. 2022The Road to Explainability is Paved with Bias: Measuring the Fairness of Explanationscited by 41
  11. 2024How Does Quantization Affect Multilingual LLMs?cited by 41
  12. 2021Your fairness may vary: Pretrained language model fairness in toxic text classificationcited by 38
  13. 2021Beyond Preserved Accuracy: Evaluating Loyalty and Robustness of BERT Compressioncited by 38
  14. 2022Sparse Random Networks for Communication-Efficient Federated Learningcited by 32
  15. 2022Pruning's Effect on Generalization Through the Lens of Training and Regularizationcited by 31
  16. 2022FairGRAPE: Fairness-aware GRAdient Pruning mEthod for Face Attribute Classificationcited by 30
  17. 2021The Rich Get Richer: Disparate Impact of Semi-Supervised Learningcited by 28
  18. 2024Accuracy is Not All You Needcited by 27
  19. 2022Metadata Archaeology: Unearthing Data Subsets by Leveraging Training Dynamicscited by 26
  20. 2023On The Impact of Machine Learning Randomness on Group Fairnesscited by 26
  21. 2024PromptKD: Distilling Student-Friendly Knowledge for Generative Language Models via Prompt Tuningcited by 24
  22. 2021An Information-Theoretic Justification for Model Pruningcited by 23
  23. 2021Hi, my name is Martha: Using names to measure and mitigate bias in generative dialogue modelscited by 23
  24. 2021Avoiding Inference Heuristics in Few-shot Prompt-based Finetuningcited by 23
  25. 2023Visual Prompting Upgrades Neural Network Sparsification: A Data-Model Perspectivecited by 23
  26. 2023Model Compression in Practice: Lessons Learned from Practitioners Creating On-device Machine Learning Experiencescited by 21
  27. 2022Bias in Automated Speaker Recognitioncited by 20
  28. 2022Train Flat, Then Compress: Sharpness-Aware Minimization Learns More Compressible Modelscited by 20
  29. 2023Sparse Model Soups: A Recipe for Improved Pruning via Model Averagingcited by 20
  30. 2021Scaling Up Exact Neural Network Compression by ReLU Stabilitycited by 19
  31. 2021A Tale Of Two Long Tailscited by 19
  32. 2021NxMTransformer: Semi-Structured Sparsification for Natural Language Understanding via ADMMcited by 19
  33. 2023Understanding the Effect of Model Compression on Social Bias in Large Language Modelscited by 19
  34. 2022Can Model Compression Improve NLP Fairnesscited by 18
  35. 2025Scaling Pre-training to One Hundred Billion Data for Vision Language Modelscited by 18
  36. 2021LMdiff: A Visual Diff Tool to Compare Language Modelscited by 17
  37. 2022Towards Fair Federated Recommendation Learning: Characterizing the Inter-Dependence of System and Data Heterogeneitycited by 17
  38. 2024On the Limitations of Compute Thresholds as a Governance Strategycited by 17

50 most cited of 162 All 162 (JSON)