Transfer learning
Reusing what a model learned on one task to jump-start a related one.
Think of it like
Like a guitarist picking up the ukulele fast — the fingers already know most of what they need.
Example
A model pretrained on general text is fine-tuned on legal documents, reaching good accuracy with far less legal data.
How it actually works
Transfer learning takes representations learned on a large, general task and adapts them to a new, often smaller task, saving data and compute. It’s the whole logic of the pretrain-then-fine-tune era: learn broadly once, specialize cheaply many times. It works because low-level features (grammar, edges, structure) generalize; it can falter when source and target domains are too dissimilar.
For product teams
The reason you rarely start from scratch — borrow a strong base and specialize for a fraction of the cost.
For engineers
Adapt representations from a source task to a target task; the foundation of the pretrain→fine-tune paradigm.
Related
- Fine-tuning — Underlies fine-tuning.
- Pretraining — Builds on pretraining.
- Multitask learning — Related to multitask learning.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome