machine unlearning

A | B | C | D | E | F | G | H | I | J | K | L | M | N | O | P | Q | R | S | T | U | V | W | X | Y | Z

Definitions:

A technique that involves selectively removing the influences of specific training data points from a trained machine learning model, such as to remove unwanted capabilities or knowledge in a foundation model, or to enable a user to request the removal of their records from a model. Efficient approximate unlearning techniques may not require retraining the ML model from scratch.
Sources:
NIST AI 100-2e2025

About

See the identified Source document to understand each term-definition pair in its proper context.

Send inquiries about terminology to the Source's authors; NIST publications will usually include a contact email for that Source.

For other inquiries, such as comments about the Glossary's presentation and functionality, use this link.

See the Glossary homepage for more guidance.