Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 132 of the 132 references it cites were found in a registry or an index; 140 of the 233 sources its claims cite were checked for support.

Machine Unlearning: A Survey

Open access
Xu, Heng, Zhu, Tianqing, Zhang, Lefeng, Zhou, Wanlei, Yu, Philip S. · 2024 · ACM Computing Surveys

Abstract

This survey captures the key concepts of unlearning techniques and highlights outstanding issues with unlearning techniques, along with feasible directions for new research opportunities.

Overview

CoChat Summary

TL;DR

This survey reviews machine unlearning, proposing a taxonomy that divides existing techniques into data reorganization and model manipulation, with verification mechanisms reviewed alongside each category. It summarizes advantages, limitations, and open issues, including universality, security, verification, and applications.

Key findings

  • The paper proposes a taxonomy of unlearning methods: data reorganization (data obfuscation, data pruning, data replacement) and model manipulation (model shifting, model pruning, model replacement). Verification methods are categorized as empirical (retraining-based, attack-based, relearn time-based, accuracy-based) or theoretical (theory-based, information bound-based).
  • Data reorganization schemes often retain intermediate parameters and use the original training dataset; they can support complex models and strong unlearning but may affect consistency and accuracy and require additional storage. Data obfuscation seldom completely unlearns information; data pruning can reduce accuracy as sub-models train on partial data; data replacement struggles to retain all original dataset information.
  • Model manipulation schemes directly adjust model parameters, reducing intermediate storage, but most focus on simple models or special assumptions; removing information from standard deep networks remains an open problem. They may affect consistency or prediction accuracy, and theory-based or information-bound verification is more convincing but complex and limited to specified models.
  • Open questions include universality of unlearning solutions, security risks where unlearning can increase privacy leakage, user-friendly verification, and broader applications. Future directions include information synchronization, federated unlearning, disturbance techniques, feature-based unlearning, and game-theory-based balance.

Methods

  • Design: a comprehensive survey that proposes a novel taxonomy based on unlearning rationale and strategy, summarizes state-of-the-art methods, reviews verification methods, and discusses open issues and future directions.
  • Data or sample: existing machine unlearning literature; the paper does not state a systematic search protocol, inclusion criteria, or quantitative synthesis.
  • Analysis: qualitative classification and comparison across dimensions such as whether original training data is required, whether intermediate data is cached, supported classes and models, unlearning targets, and verification methods.

Limitations

  • The paper states limitations of existing unlearning schemes: exact unlearning can only be achieved for simple and well-structured models; many schemes lack theoretical verification; data reorganization can incur storage and accuracy costs; model manipulation is often limited to simple models or special assumptions; and many verification methods are one-sided or not user-friendly.
  • The paper does not explicitly state limitations of the survey itself. Evident from the text: it does not report a systematic review protocol or quantitative meta-analysis, and its coverage reflects the literature cited up to the publication date.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3603620

Cites

132 of 132 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

104 Valid · 10 Not valid · 55 Not checked · 91 Own results and interpretations

86 more claims · 36 with their cited sources checked

Create a free account to see all 260 claims and to verify the 12 references still open.

See all claims — free account
  • Probably not a substantial claim. 128
  • It may state two separate claims as one. 68
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

3 of 3

Citations

Cites

Show all 122 references

Research works printed but not linked: 5

  1. Sebastian Schelter. 2020. “Amnesia’—Towards machine learning models that can forget user data very fast. In 10th Conference on Innovative Data Systems Research. Retrieved from http://cidrdb.org/cidr2020/papers/p32-schelter-cidr20.pdf.
  2. Pei-Hung Chen, Wei Wei, Cho-Jui Hsieh, and Bo Dai. 2021. Overcoming catastrophic forgetting by Bayesian generative regularization. In 38th International Conference on Machine Learning(Proceedings of Machine Learning Research, Vol. 139). PMLR, 1760–1770. Retrieved from http://proceedings.mlr.press/v139/chen21v.html.
  3. Zachary Izzo, Mary Anne Smart, Kamalika Chaudhuri, and James Zou. 2021. Approximate data deletion from machine learning models. In 24th International Conference on Artificial Intelligence and Statistics(Proceedings of Machine Learning Research, Vol. 130). PMLR, 2008–2016. Retrieved from http://proceedings.mlr.press/v130/izzo21a.html.
  4. Yinjun Wu, Edgar Dobriban, and Susan B. Davidson. 2020. DeltaGrad: Rapid retraining of machine learning models. In 37th International Conference on Machine Learning(Proceedings of Machine Learning Research, Vol. 119). PMLR, 10355–10366. Retrieved from http://proceedings.mlr.press/v119/wu20b.html.
  5. 10.5555/1953048.2021036 · doi:10.5555/1953048.2021036

Cited by

See all 164

Metadata

Authors
Xu, H., Zhu, T., Zhang, L., Zhou, W., & Yu, P. S.
Affiliations
  • University of Technology Sydney (AU)
  • City University of Macau (MO)
  • University of Illinois Chicago (US)
Venue
ACM Computing Surveys
Published
2023-08-28
Volume
56
Issue
1
Pages
1-36
ISSN
  • 0360-0300 (ISSN-L)
  • 1557-7341
OpenAlex
W4379652910
Cited by
164
References
122
Citations by year
2026: 35 · 2025: 77 · 2024: 46 · 2023: 6
FWCI
14.51
Type
journal article
Language
en
Open access
Locations
  • ACM Computing Surveys (publishedVersion, other-oa)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • arXiv (Cornell University) (cc-by)
Funders
  • Australian Research Council Discovery (DP200100946 and DP230100246)
  • NSF (III-1763325, III-1909323, III-2106758, and SaTC-1930941)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
  • Physical Sciences › Computer Science › Computer Networks and Communications › Cryptography and Data Security
  • Physical Sciences › Computer Science › Information Systems › Cloud Data Security Solutions
Keywords
machine unlearning, right to be forgotten, privacy, model retraining, privacy-preserving machine learning

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-12, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-12 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Machine Unlearning: A Survey. https://app.cochat.ai/papers/10.1145/3603620
The paper
Xu, H., Zhu, T., Zhang, L., Zhou, W., & Yu, P. S. (2024). Machine Unlearning: A Survey. ACM Computing Surveys. https://doi.org/10.1145/3603620

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.