Machine Unlearning
Techniques that make a trained model behave as if specific data had never been in its training set without a full retrain. Exact methods (SISA, Sharded-Isolated-Sliced-Aggregated, Bourtoule et al. 2021) retrain only the affected shard and are available in production ML platforms that support them. Approximate methods statistically suppress deleted data's influence without retraining; they offer strong evidence of forgetting but not yet an ironclad mathematical guarantee, and verification remains an open research problem in 2026.
Defined in 3 GAGE programs, which carry 8 distinct definitions of it. The wording above is taught in AI Literacy & Professional Conduct.
How each discipline defines it
The same term does different work depending on who is using it. These are the definitions as each program teaches them, unedited.
Techniques that make a trained model behave as if specific data had never been in its training set without a full retrain. Exact methods (SISA, Sharded-Isolated-Sliced-Aggregated, Bourtoule et al. 2021) retrain only the affected shard and are available in production ML platforms that support them. Approximate methods statistically suppress deleted data's influence without retraining; they offer strong evidence of forgetting but not yet an ironclad mathematical guarantee, and verification remains an open research problem in 2026.
An emerging class of techniques attempting to remove the influence of specific training examples from an already-trained model without full retraining. As of 2026, the field lacks a standardized approach and any formal, verifiable guarantee that unlearning has fully succeeded; treat claims of successful unlearning as unproven unless independently verified.
The emerging and, as of 2026, not-yet-reliable field of removing the influence of specific training data from an already-trained model without full retraining.
A body of research techniques attempting to remove a specific training example's influence from an already-trained model's weights without a full retrain from scratch. As of August 2026, an emerging field without standardized definitions of success or reliable, widely accepted verification methods; not a production-grade guaranteed deletion capability.
An emerging class of techniques attempting to remove the influence of specific training records from an already-trained model without retraining from scratch; as of August 2026, no standardized method offers a formally guaranteed or reliably verifiable removal.
Where it is taught
The exact lessons this term appears in. The first 7 topics of every program are free with a free account.
- Privacy and Data Rights in AI Systems · Ethical and Responsible AI and Operational Governance, AI Literacy & Professional Conduct
- Retention versus the model that memorized: deleting data a model already learned from · Consent, Purpose, and the Law of Data, AI Data Governance: The Data Chair
- Membership inference and extraction: what a model reveals about its training data · Poison, Leaks, and the Adversary, AI Data Governance: The Data Chair
- The deletion decision: the terabytes your organization should destroy this quarter, defended · The Money of Data, AI Data Governance: The Data Chair
- The deletion that was not: personal data found in a model trained last year · Data Incidents, AI Data Governance: The Data Chair
- Provenance under attack: your content credentials stripped, spoofed, and laundered, and the countermeasures that survive · Adversarial Data Governance, AI Data Governance: The Data Chair
- Machine unlearning: the emerging answer to deleting what a model learned · The Frontier Discipline, AI Data Governance: The Data Chair
- The retention decision: what you must keep, what you must destroy, and proving both · Data Reality, AI Governance: Applied Mastery
Terms it appears with
Not an alphabetical neighbourhood: these are the terms taught in the same lessons, ranked by how often they appear together.