AI Governance Engineer
GitHub

Machine unlearning

By Updated

Techniques that remove a training record's influence from a model without full retraining. Exact methods retrain an affected shard 1; approximate methods adjust weights and are hard to verify, so an unlearning claim is tested with membership-inference or extraction evals.

Developed in
ch. 19, Suppression, retraining and unlearning
ch. 05, Pattern: Rights Requests Against Models
Chapters
ch. 05, Patterns · ch. 19, Privacy & AI
Contrast with
Output suppression
Source
1 numbered reference, listed below

Where it is used

One chapter of the Body of Knowledge uses the term. Each link opens the first section that does.

Patterns that use this term

One pattern page uses the term.

  • Fulfilment record

    The per-request record of how a data-subject request was honoured wherever the person's data sits, from source systems, snapshots, retrieval indexes, logs and…

  • Evals

    Automated tests of a model's or agent's behaviour (capability, safety and adversarial), run as controls, not as one-off research.

Sources

  1. [1] "Machine Unlearning" (Bourtoule et al.; SISA training; arXiv 1912.03817). arXiv. 2019-12-09. https://arxiv.org/abs/1912.03817 (verified: primary)

Definitions of legal terms paraphrase the cited text, which governs. Dated statements are as of .

Cite this term

García Aibar, J. (2026). Machine unlearning. In AI Governance Engineering: The Thesis & Body of Knowledge (v0.5.0), Glossary. https://doi.org/10.5281/zenodo.22956197. https://aigovernanceengineer.com/glossary/machine-unlearning. CC BY 4.0

BibTeX

@misc{aige2026machineunlearning,
  author  = {Jorge García Aibar},
  title   = {{Machine unlearning}},
  note    = {Glossary, AI Governance Engineering: The Thesis \& Body of Knowledge, version 0.5.0},
  year    = {2026},
  doi     = {10.5281/zenodo.22956197},
  url     = {https://aigovernanceengineer.com/glossary/machine-unlearning}
}