mlsanityData to 7 Oct 2026

Understanding Black-box Predictions via Influence Functions

arXiv 1703.04730 ↗First posted 14 Mar 2017stat.ML cs.AI cs.LGcited by 1,714

Not on the map. Why?

Cited by 257 in LLM safety and security, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

How can we explain the predictions of a black-box model? In this paper, we use influence functions -- a classic technique from robust statistics -- to trace a model's prediction through the learning algorithm and back to its training data, thereby identifying training points most responsible for a given prediction. To scale up influence functions to modern machine learning settings, we develop a simple, efficient implementation that requires only oracle access to gradients and Hessian-vector products. We show that even on non-convex and non-differentiable models where the theory breaks down, approximations to influence functions can still provide valuable information. On linear models and convolutional neural networks, we demonstrate that influence functions are useful for multiple purposes: understanding model behavior, debugging models, detecting dataset errors, and even creating visually-indistinguishable training-set attacks.

Key facts. Cited by 1,714 papers in the map (data through 7 Oct 2026). First posted on arXiv 14 Mar 2017.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2018: 1; 2019: 0; 2020: 0; 2021: 211; 2022: 220; 2023: 254; 2024: 320; 2025: 372; 2026: 336.

1,714 papers.

Sort:

By area:

  1. 2023CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Societycited by 1,348
  2. 2023DataComp: In search of the next generation of multimodal datasetscited by 635
  3. 2021Deep Learning on a Data Diet: Finding Important Examples Early in Trainingcited by 529
  4. 2024LESS: Selecting Influential Data for Targeted Instruction Tuningcited by 514
  5. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearningcited by 484
  6. 2021Towards Out-Of-Distribution Generalization: A Surveycited by 431
  7. 2023SalUn: Empowering Machine Unlearning via Gradient-based Weight Saliency in Both Image Classification and Generationcited by 376
  8. 2024LMMs-Eval: Reality Check on the Evaluation of Large Multimodal Modelscited by 344
  9. 2023Explainability for Large Language Models: A Surveycited by 331
  10. 2022A Survey of Machine Unlearningcited by 298
  11. 2021Understanding Dataset Difficulty with V-Usable Informationcited by 288
  12. 2023Large Language Model Unlearningcited by 288
38 more
  1. 2023TRAK: Attributing Model Behavior at Scalecited by 273
  2. 2024Rethinking Machine Unlearning for Large Language Modelscited by 247
  3. 2021Adversarial Neuron Pruning Purifies Backdoored Deep Modelscited by 230
  4. 2021Machine Unlearning of Features and Labelscited by 223
  5. 2021Fast Yet Effective Machine Unlearningcited by 209
  6. 2021Trustworthy AI: From Principles to Practicescited by 206
  7. 2023Model Sparsity Can Simplify Machine Unlearningcited by 206
  8. 2023To Generate or Not? Safety-Driven Unlearned Diffusion Models Are Still Easy To Generate Unsafe Images ... For Nowcited by 204
  9. 2021Unlearnable Examples: Making Personal Data Unexploitablecited by 188
  10. 2022Datamodels: Predicting Predictions from Training Datacited by 186
  11. 2024RegMix: Data Mixture as Regression for Language Model Pre-trainingcited by 172
  12. 2021On the Necessity of Auditable Algorithmic Definitions for Machine Unlearningcited by 164
  13. 2021Beta Shapley: a Unified and Noise-reduced Data Valuation Framework for Machine Learningcited by 162
  14. 2021Adversarial Unlearning of Backdoors via Implicit Hypergradientcited by 160
  15. 2022A Review on Language Models as Knowledge Basescited by 156
  16. 2021Rethinking the Backdoor Attacks' Triggers: A Frequency Perspectivecited by 155
  17. 2021Post-hoc Interpretability for Neural NLP: A Surveycited by 155
  18. 2021Counterfactual Memorization in Neural Language Modelscited by 155
  19. 2021Poisoning and Backdooring Contrastive Learningcited by 149
  20. 2021Explainable artificial intelligence (XAI) in deep learning-based medical image analysiscited by 147
  21. 2022If Influence Functions are the Answer, Then What is the Question?cited by 142
  22. 2024Benchmark Data Contamination of Large Language Models: A Surveycited by 138
  23. 2021Benchmarking and Survey of Explanation Methods for Black Box Modelscited by 136
  24. 2022Training Data Influence Analysis and Estimation: A Surveycited by 134
  25. 2022Dataset Pruning: Reducing Training Data by Examining Generalization Influencecited by 132
  26. 2024Tamper-Resistant Safeguards for Open-Weight LLMscited by 132
  27. 2021Scaling Up Influence Functionscited by 130
  28. 2022Towards Faithful Model Explanation in NLP: A Surveycited by 129
  29. 2022Data Banzhaf: A Robust Data Valuation Framework for Machine Learningcited by 126
  30. 2022Explanations from Large Language Models Make Small Reasoners Bettercited by 123
  31. 2023DataInf: Efficiently Estimating Data Influence in LoRA-tuned LLMs and Diffusion Modelscited by 121
  32. 2023Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Modelscited by 120
  33. 2023AI Transparency in the Age of LLMs: A Human-Centered Research Roadmapcited by 119
  34. 2021Interpretable Deep Learning: Interpretation, Interpretability, Trustworthiness, and Beyondcited by 116
  35. 2024Interpreting CLIP with Sparse Linear Concept Embeddings (SpLiCE)cited by 116
  36. 2021Adversarial Examples Make Strong Poisonscited by 115
  37. 2021Salient ImageNet: How to discover spurious features in Deep Learning?cited by 114
  38. 2023Foundation Models and Fair Usecited by 112

50 most cited of 1,714 All 1,714 (JSON)