AI Glossary Term
What does “Domain Adaptation” mean?
Definition
Methods that transfer a model trained in one domain to perform better in another domain.
How Domain Adaptation works
Example
A speech recognizer trained on clear studio recordings performs poorly on noisy call-center audio. By adapting it with a small amount of call-center audio, accuracy improves without training from scratch.
Why it matters
Real data rarely matches training data exactly. Adaptation is a cost-effective way to cover that gap.
Common questions
How is it different from fine-tuning?
Fine-tuning is one common way to do domain adaptation. The term also covers methods that use unlabeled target data.
What if I have no labels in the new domain?
Unsupervised domain adaptation tries to align features between domains using only unlabeled target examples.
How do I know I need it?
A drop in accuracy when moving from test data to real-world data is a common sign of domain shift.