mlsanityData to 7 Oct 2026

Backward Compatibility During Data Updates by Weight Interpolation

arXiv 2301.10546 ↗First posted 25 Jan 2023cs.LG cs.CLcited by 7

Not on the map. Why?

Abstract (arXiv)

Backward compatibility of model predictions is a desired property when updating a machine learning driven application. It allows to seamlessly improve the underlying model without introducing regression bugs. In classification tasks these bugs occur in the form of negative flips. This means an instance that was correctly classified by the old model is now classified incorrectly by the updated model. This has direct negative impact on the user experience of such systems e.g. a frequently used voice assistant query is suddenly misclassified. A common reason to update the model is when new training data becomes available and needs to be incorporated. Simply retraining the model with the updated data introduces the unwanted negative flips. We study the problem of regression during data updates and propose Backward Compatible Weight Interpolation (BCWI). This method interpolates between the weights of the old and new model and we show in extensive experiments that it reduces negative flips without sacrificing the improved accuracy of the new model. BCWI is straight forward to implement and does not increase inference cost. We also explore the use of importance weighting during interpolation and averaging the weights of multiple new models in order to further reduce negative flips.

Key facts. Cited by 7 papers in the map (data through 7 Oct 2026). Builds on 34 of its 46 references that are in the map. First posted on arXiv 25 Jan 2023.

Builds on

34 of 46 references.

References · 34

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  3. 2015Distilling the Knowledge in a Neural Networkcited by 10,076
  4. 2016Overcoming catastrophic forgetting in neural networkscited by 4,718
  5. 2016iCaRL: Incremental Classifier and Representation Learningcited by 2,313
  6. 2015Character-level Convolutional Networks for Text Classificationcited by 2,293
  7. 2017Gradient Episodic Memory for Continual Learningcited by 1,871
  8. 2022Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference timecited by 1,366
  9. 2018Continual Lifelong Learning with Neural Networks: A Reviewcited by 1,313
  10. 2021BitFit: Simple Parameter-efficient Fine-tuning for Transformer-based Masked Language-modelscited by 1,234
  11. 2018Averaging Weights Leads to Wider Optima and Better Generalizationcited by 1,194
  12. 2022Few-Shot Parameter-Efficient Fine-Tuning is Better and Cheaper than In-Context Learningcited by 886
22 more
  1. 2021Robust fine-tuning of zero-shot modelscited by 785
  2. 2018An Empirical Study of Example Forgetting during Deep Neural Network Learningcited by 669
  3. 2021Merging Models with Fisher-Weighted Averagingcited by 613
  4. 2021Editing Factual Knowledge in Language Modelscited by 548
  5. 2021Fast Model Editing at Scalecited by 491
  6. 2022Git Re-Basin: Merging Models modulo Permutation Symmetriescited by 460
  7. 2020Efficient Intent Detection with Dual Sentence Encoderscited by 419
  8. 2020On the Stability of Fine-tuning BERT: Misconceptions, Explanations, and Strong Baselinescited by 254
  9. 2022Patching open-vocabulary models by interpolating weightscited by 187
  10. 2022Diverse Weight Averaging for Out-of-Distribution Generalizationcited by 148
  11. 2022MASSIVE: A 1M-Example Multilingual Natural Language Understanding Dataset with 51 Typologically-Diverse Languagescited by 142
  12. 2019Mixout: Effective Regularization to Finetune Large-scale Pretrained Language Modelscited by 141
  13. 2020Towards Backward-Compatible Representation Learningcited by 74
  14. 2020Positive-Congruent Training: Towards Regression-Free Model Updatescited by 55
  15. 2020An Empirical Analysis of Backward Compatibility in Machine Learning Systemscited by 38
  16. 2017Neural Domain Adaptation for Biomedical Question Answeringcited by 30
  17. 2021Regression Bugs Are In Your Model! Measuring, Reducing and Analyzing Regressions In NLP Model Updatescited by 13
  18. 2022An Introduction to Lifelong Supervised Learningcited by 12
  19. 2022Measuring and Reducing Model Update Regression in Structured Prediction for NLPcited by 11
  20. 2022Model Stability with Continuous Data Updatescited by 9
  21. 2022ELODI: Ensemble Logit Difference Inhibition for Positive-Congruent Trainingcited by 7
  22. 2022Reducing Model Jitter: Stable Re-training of Semantic Parsers in Production Environmentscited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 1; 2024: 2; 2025: 1; 2026: 3.

7 papers.

  1. 2024Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunitiescited by 247
  2. 2023Improving Prediction Backward-Compatiblility in NLP Model Upgrade with Gated Fusioncited by 6
  3. 2024DEM: Distribution Edited Model for Training with Mixed Data Distributionscited by 3
  4. 2026System 1&2 Synergy via Dynamic Model Interpolationcited by 3
  5. 2025Regression-aware Continual Learning for Android Malware Detectioncited by 2
  6. 2026Same Agent, Different Answers: A Repeat-Aware Audit of Corpus-Induced Answer Churn in Retrieval-Augmented QAcited by 1
  7. 2026Learn Here, Move Less Elsewhere: Input-Conditioned Plasticity from Retained-Domain Activation Atlasescited by 0