Domain Adaptation
Taking a model trained on one kind of data and adjusting it to work well on a different kind.
Think of it like
A driver who knows city streets taking a short course before driving a truck on mountain roads.
Example
A general chat model is adapted on medical notes so it handles clinical shorthand instead of stumbling on it.
How it actually works
A model performs best on data resembling what it trained on; move it to a new domain — legal, medical, a new language style — and accuracy drops. Domain adaptation closes that gap with additional training on target-domain data, sometimes without labels. The subtlety is doing it without erasing the general capabilities that made the base model useful.
For product teams
How you turn a general model into a specialist for your industry’s vocabulary and quirks.
For engineers
Adjusting a model to a target data distribution that differs from its source training distribution.
Related
- Transfer learning — A form of transfer learning.
- Fine-tuning — Often done via fine-tuning.
- Continued pretraining — Or continued pretraining.
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