What is Transfer Learning

Definition

Transfer learning is a machine-learning technique where a model developed for one task is reused as the starting point for a related task, leveraging knowledge already learned so the new model needs far less data and training time to perform well.

« Back to Glossary Index

Reuses knowledge from a pre-trained model, cutting data and compute needs
Achieves strong performance on tasks with limited labelled data
Speeds development by starting from a capable base rather than scratch
Makes advanced models practical for narrow, data-scarce problems

Real World Example

A startup with only a few thousand labelled medical images fine-tunes a model pre-trained on millions of general images, achieving accurate diagnosis support that would have been impossible training from scratch on its small dataset.

FAQs

Why is transfer learning useful?

It lets you build accurate models with limited data and compute by reusing what a model already learned on a related task.

How does transfer learning work?

A pre-trained model's learned representations are reused, often with later layers fine-tuned on the new task's data.

When does transfer learning fail?

When the source and target tasks are too dissimilar, transferred knowledge may not help and can even hurt performance.