Transfer Learning

Level 1

When AI uses what it learned from one thing to help with something new — like how learning to ride a bike helps you learn a motorcycle.

Level 2

The idea that an AI trained on one task can apply that knowledge to a different task — so you don't have to start from scratch every time.

Level 3

A technique where knowledge learned from one task or domain is applied to a different but related task, reducing the data and compute needed for the new task.

Level 4

Leveraging representations learned during pre-training on a source task to improve performance on a target task — the fundamental paradigm behind foundation models and their downstream adaptations.

Level 5

The exploitation of shared structure between source and target domains via learned representations — formalized as minimizing target risk under domain shift, with theoretical bounds governed by the divergence between source and target distributions.

Sources

• NIST Trustworthy and Responsible AI Resource Center: Glossary — National Institute of Standards and Technology. Accessed 2026-07-20.