# An Empirical Analysis of Backward Compatibility in Machine Learning Systems

arXiv: https://arxiv.org/abs/2008.04572 · HTML: https://mlsanity.com/paper/2008.04572/

## Key facts

- Cited by 38 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 11 Aug 2020.

## Built on by (most cited first, 30 of 38)

- [Capabilities of GPT-4 on Medical Challenge Problems](https://mlsanity.com/paper/2303.13375/) (2023)
- [Class-Incremental Learning: A Survey](https://mlsanity.com/paper/2302.03648/) (2023)
- [Forward Compatible Few-Shot Class-Incremental Learning](https://mlsanity.com/paper/2203.06953/) (2022)
- [Human-Algorithm Collaboration: Achieving Complementarity and Avoiding Unfairness](https://mlsanity.com/paper/2202.08821/) (2022)
- [Machine Learning for Microcontroller-Class Hardware: A Review](https://mlsanity.com/paper/2205.14550/) (2022)
- [MUSCLE: A Model Update Strategy for Compatible LLM Evolution](https://mlsanity.com/paper/2407.09435/) (2024)
- [Backward-Compatible Prediction Updates: A Probabilistic Approach](https://mlsanity.com/paper/2107.01057/) (2021)
- [CoReS: Compatible Representations via Stationarity](https://mlsanity.com/paper/2111.07632/) (2021)
- [Boundary-aware Backward-Compatible Representation via Adversarial Learning in Image Retrieval](https://mlsanity.com/paper/2305.02610/) (2023)
- [AI Model Disgorgement: Methods and Choices](https://mlsanity.com/paper/2304.03545/) (2023)
- [NeuRecover: Regression-Controlled Repair of Deep Neural Networks with Training History](https://mlsanity.com/paper/2203.00191/) (2022)
- [Efficient Construction of Model Family through Progressive Training Using Model Expansion](https://mlsanity.com/paper/2504.00623/) (2025)
- [Hot-Refresh Model Upgrades with Regression-Alleviating Compatible Training in Image Retrieval](https://mlsanity.com/paper/2201.09724/) (2022)
- [ELODI: Ensemble Logit Difference Inhibition for Positive-Congruent Training](https://mlsanity.com/paper/2205.06265/) (2022)
- [Backward Compatibility During Data Updates by Weight Interpolation](https://mlsanity.com/paper/2301.10546/) (2023)
- [FastFill: Efficient Compatible Model Update](https://mlsanity.com/paper/2303.04766/) (2023)
- [Optimistic Verifiable Training by Controlling Hardware Nondeterminism](https://mlsanity.com/paper/2403.09603/) (2024)
- [Fixes That Fail: Self-Defeating Improvements in Machine-Learning Systems](https://mlsanity.com/paper/2103.11766/) (2021)
- [Understanding User Intent Modeling for Conversational Recommender Systems: A Systematic Literature Review](https://mlsanity.com/paper/2308.08496/) (2023)
- [Robustness-Congruent Adversarial Training for Secure Machine Learning Model Updates](https://mlsanity.com/paper/2402.17390/) (2024)
- [Distributionally Robust Group Backwards Compatibility](https://mlsanity.com/paper/2112.10290/) (2021)
- [A Comprehensive Study of Deep Learning Model Fixing Approaches](https://mlsanity.com/paper/2512.23745/) (2025)
- [Pay Only for Disagreement: Certified No-Regression Verdicts for Model Updates with Matching Label-Complexity Bounds](https://mlsanity.com/paper/2609.17560/) (2026)
- [Hidden Heterogeneity: When to Choose Similarity-Based Calibration](https://mlsanity.com/paper/2202.01840/) (2022)
- [Maintaining Stability and Plasticity for Predictive Churn Reduction](https://mlsanity.com/paper/2305.04135/) (2023)
- [Updating Clinical Risk Stratification Models Using Rank-Based Compatibility: Approaches for Evaluating and Optimizing Clinician-Model Team Performance](https://mlsanity.com/paper/2308.05619/) (2023)
- [Backward Compatibility in Attributive Explanation and Enhanced Model Training Method](https://mlsanity.com/paper/2408.02298/) (2024)
- [An Experience Report on Regression-Free Repair of Deep Neural Network Model](https://mlsanity.com/paper/2503.07079/) (2025)
- ["Who experiences large model decay and why?" A Hierarchical Framework for Diagnosing Heterogeneous Performance Drift](https://mlsanity.com/paper/2506.00756/) (2025)
- [Practical Insights of Repairing Model Problems on Image Classification](https://mlsanity.com/paper/2205.07116/) (2022)

## Abstract (arXiv)

In many applications of machine learning (ML), updates are performed with the goal of enhancing model performance. However, current practices for updating models rely solely on isolated, aggregate performance analyses, overlooking important dependencies, expectations, and needs in real-world deployments. We consider how updates, intended to improve ML models, can introduce new errors that can significantly affect downstream systems and users. For example, updates in models used in cloud-based classification services, such as image recognition, can cause unexpected erroneous behavior in systems that make calls to the services. Prior work has shown the importance of "backward compatibility" for maintaining human trust. We study challenges with backward compatibility across different ML architectures and datasets, focusing on common settings including data shifts with structured noise and ML employed in inferential pipelines. Our results show that (i) compatibility issues arise even without data shift due to optimization stochasticity, (ii) training on large-scale noisy datasets often results in significant decreases in backward compatibility even when model accuracy increases, and (iii) distributions of incompatible points align with noise bias, motivating the need for compatibility aware de-noising and robustness methods.