mlsanityData to 14 Oct 2026

Causality-based Neural Network Repair

arXiv 2204.09274 ↗First posted 20 Apr 2022cs.SE cs.AI cs.LGcited by 47

On the map: LLM safety and security → Backdoor and data poisoning attacks

Abstract (arXiv)

Neural networks have had discernible achievements in a wide range of applications. The wide-spread adoption also raises the concern of their dependability and reliability. Similar to traditional decision-making programs, neural networks can have defects that need to be repaired. The defects may cause unsafe behaviors, raise security concerns or unjust societal impacts. In this work, we address the problem of repairing a neural network for desirable properties such as fairness and the absence of backdoor. The goal is to construct a neural network that satisfies the property by (minimally) adjusting the given neural network's parameters (i.e., weights). Specifically, we propose CARE (CAusality-based REpair), a causality-based neural network repair technique that 1) performs causality-based fault localization to identify the `guilty' neurons and 2) optimizes the parameters of the identified neurons to reduce the misbehavior. We have empirically evaluated CARE on various tasks such as backdoor removal, neural network repair for fairness and safety properties. Our experiment results show that CARE is able to repair all neural networks efficiently and effectively. For fairness repair tasks, CARE successfully improves fairness by 61.91% on average. For backdoor removal tasks, CARE reduces the attack success rate from over 98% to less than 1%. For safety property repair tasks, CARE reduces the property violation rate to less than 1%. Results also show that thanks to the causality-based fault localization, CARE's repair focuses on the misbehavior and preserves the accuracy of the neural networks.

Key facts. Cited by 47 papers in the map (data through 14 Oct 2026). Placed in Backdoor and data poisoning attacks, one of 12 directions in LLM safety and security. Builds on 32 of its 70 references that are in the map. First posted on arXiv 20 Apr 2022.

Builds on

32 of 70 references.

References · 32

  1. 2017Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithmscited by 4,349
  2. 2015FaceNet: A Unified Embedding for Face Recognition and Clusteringcited by 2,849
  3. 2016End to End Learning for Self-Driving Carscited by 1,009
  4. 2017Counterfactual Fairnesscited by 804
  5. 2017Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networkscited by 744
  6. 2014Certifying and removing disparate impactcited by 715
  7. 2017Formal Verification of Piece-Wise Linear Feed-Forward Neural Networkscited by 242
  8. 2018Formal Security Analysis of Neural Networks using Symbolic Intervalscited by 185
  9. 2017CausalGAN: Learning Causal Implicit Generative Models with Adversarial Trainingcited by 152
  10. 2018Counterfactual Fairness in Text Classification through Robustnesscited by 151
  11. 2019Branch and Bound for Piecewise Linear Neural Network Verificationcited by 138
  12. 2018Automated Directed Fairness Testingcited by 78
20 more
  1. 2019Neural Network Attributions: A Causal Perspectivecited by 77
  2. 2019An Abstraction-Based Framework for Neural Network Verificationcited by 68
  3. 2018DeepMutation: Mutation Testing of Deep Learning Systemscited by 67
  4. 2018Deep Neural Network Compression for Aircraft Collision Avoidance Systemscited by 58
  5. 2020Verifying Individual Fairness in Machine Learning Modelscited by 49
  6. 2018Probabilistic Verification of Fairness Properties via Concentrationcited by 45
  7. 2021Provable Repair of Deep Neural Networkscited by 42
  8. 2019POPQORN: Quantifying Robustness of Recurrent Neural Networkscited by 37
  9. 2020Verifying Recurrent Neural Networks using Invariant Inferencecited by 26
  10. 2021NNrepair: Constraint-based Repair of Neural Network Classifierscited by 26
  11. 2018Causal Learning and Explanation of Deep Neural Networks via Autoencoded Activationscited by 23
  12. 2018Causal Testing: Finding Defects' Root Causescited by 22
  13. 2018Automated Test Generation to Detect Individual Discrimination in AI Modelscited by 21
  14. 2018Explaining Deep Learning Models using Causal Inferencecited by 18
  15. 2019Explaining Visual Models by Causal Attributioncited by 18
  16. 2021Probabilistic Verification of Neural Networks Against Group Fairnesscited by 17
  17. 2020Towards Repairing Neural Networks Correctlycited by 14
  18. 2020SOCRATES: Towards a Unified Platform for Neural Network Analysiscited by 7
  19. 2020Robustness Verification for Classifier Ensemblescited by 3
  20. 2017Optimization assisted MCMCcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 2; 2023: 16; 2024: 12; 2025: 10; 2026: 7.

47 papers.

Sort:

By area:

  1. 2022Adaptive Fairness Improvement Based on Causality Analysiscited by 17
  2. 2024Eliminating Information Leakage in Hard Concept Bottleneck Models with Supervised, Hierarchical Concept Learningcited by 17
  3. 2023Benchmarking and Explaining Large Language Model-based Code Generation: A Causality-Centric Approachcited by 15
  4. 2024LLMScan: Causal Scan for LLM Misbehavior Detectioncited by 13
  5. 2022Applications of statistical causal inference in software engineeringcited by 12
  6. 2023Causality Analysis for Evaluating the Security of Large Language Modelscited by 11
  7. 2024NeuFair: Neural Network Fairness Repair with Dropoutcited by 11
  8. 2024Machine Learning Robustness: A Primercited by 8
  9. 2023Causality-Aided Trade-off Analysis for Machine Learning Fairnesscited by 6
  10. 2024Online Safety Analysis for LLMs: a Benchmark, an Assessment, and a Path Forwardcited by 6
  11. 2023Neuron Patching: Semantic-based Neuron-level Language Model Repair for Code Generationcited by 5
  12. 2024Certified Continual Learning for Neural Network Regressioncited by 5
35 more
  1. 2023Repairing Deep Neural Networks Based on Behavior Imitationcited by 4
  2. 2023LUNA: A Model-Based Universal Analysis Framework for Large Language Modelscited by 4
  3. 2023Semantic-Based Neural Network Repaircited by 3
  4. 2023How Generalizable are Deepfake Image Detectors? An Empirical Studycited by 3
  5. 2024Causal Reasoning in Software Quality Assurance: A Systematic Reviewcited by 3
  6. 2025Causal-Guided Detoxify Backdoor Attack of Open-Weight LoRA Modelscited by 3
  7. 2023Incremental Satisfiability Modulo Theory for Verification of Deep Neural Networkscited by 2
  8. 2023FITNESS: A Causal De-correlation Approach for Mitigating Bias in Machine Learning Softwarecited by 2
  9. 2023Enabling Runtime Verification of Causal Discovery Algorithms with Automated Conditional Independence Reasoning (Extended Version)cited by 2
  10. 2025A Semantic-based Optimization Approach for Repairing LLMs: Case Study on Code Generationcited by 2
  11. 2025A Comprehensive Study of Deep Learning Model Fixing Approachescited by 2
  12. 2023Reasoning-Based Software Testingcited by 1
  13. 2023Towards Better Fairness-Utility Trade-off: A Comprehensive Measurement-Based Reinforcement Learning Frameworkcited by 1
  14. 2023TraceFL: Interpretability-Driven Debugging in Federated Learning via Neuron Provenancecited by 1
  15. 2024Inference-Time Rule Eraser: Fair Recognition via Distilling and Removing Biased Rulescited by 1
  16. 2024NeuSemSlice: Towards Effective DNN Model Maintenance via Neuron-level Semantic Slicingcited by 1
  17. 2025NeuroBreak: Unveil Internal Jailbreak Mechanisms in Large Language Modelscited by 1
  18. 2026AtPatch: Debugging Transformers via Hot-Fixing Over-Attentioncited by 1
  19. 2023Understanding the Complexity and Its Impact on Testing in ML-Enabled Systemscited by 0
  20. 2023Path Analysis for Effective Fault Localization in Deep Neural Networkscited by 0
  21. 2024Patch Synthesis for Property Repair of Deep Neural Networkscited by 0
  22. 2024Purification Of Contaminated Convolutional Neural Networks Via Robust Recovery: An Approach with Theoretical Guarantee in One-Hidden-Layer Casecited by 0
  23. 2024An Empirical Study of Fault Localisation Techniques for Deep Learningcited by 0
  24. 2025PRUNE: A Patching Based Repair Framework for Certifiable Unlearning of Neural Networkscited by 0
  25. 2025Search-Based Software Engineering and AI Foundation Models: Current Landscape and Future Roadmapcited by 0
  26. 2025AR2: Attention-Guided Repair for the Robustness of CNNs Against Common Corruptionscited by 0
  27. 2025An Empirical Study of Sample Selection Strategies for Large Language Model Repaircited by 0
  28. 2025Provable Repair of Deep Neural Network Defects by Preimage Synthesis and Property Refinementcited by 0
  29. 2025Fairness Is Not Just Ethical: Performance Trade-Off via Data Correlation Tuning to Mitigate Bias in ML Softwarecited by 0
  30. 2026ProToken: Token-Level Attribution for Federated Large Language Modelscited by 0
  31. 2026QuanForge: A Mutation Testing Framework for Quantum Neural Networkscited by 0
  32. 2026Provable Fairness Repair for Deep Neural Networkscited by 0
  33. 2026RepTran: Search-Based Repair of Transformer Modelscited by 0
  34. 2026Fairness Invariants: A Relational Approach to Explaining and Mitigating Fairness Bugscited by 0
  35. 2026FORGE: Verification-Gated Behavioral Repair for Generative Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Causality-Driven Neural Network Repair: Challenges and Opportunities
  2. 2026Shapley-Guided Neural Repair Approach via Derivative-Free Optimization
  3. 2023QNNRepair: Quantized Neural Network Repair
  4. 2021Adversarial Neuron Pruning Purifies Backdoored Deep Models
  5. 2023Backdoor Defense via Deconfounded Representation Learning
  6. 2024Fusing Pruned and Backdoored Models: Optimal Transport-based Data-free Backdoor Mitigation
  7. 2024Rethinking Pruning for Backdoor Mitigation: An Optimization Perspective
  8. 2025Combining Causal Models for More Accurate Abstractions of Neural Networks