Skip to content

What is Transfer Learning?

Transfer learning is a machine learning method that reuses a model trained on one task as the starting point for a different but related task.

Instead of training a model from random weights on a new dataset, transfer learning starts with weights already learned from a large source dataset. This captures general features that can be useful elsewhere.

The pre-trained model is then adapted, often by fine-tuning some or all of its layers on the smaller target dataset. Early layers usually stay frozen because they hold low-level patterns, while later layers are updated for the new task.

This approach reduces training time, lowers data requirements, and often yields better performance when labeled data for the target task is scarce.

Example

A model pre-trained on millions of everyday photos (ImageNet) can be fine-tuned with just a few hundred labeled X-ray images to detect pneumonia, achieving high accuracy without needing a massive medical dataset.

Why it matters

Transfer learning makes powerful AI practical in domains where collecting large labeled datasets is expensive or impossible, dramatically speeding up development and deployment of new models.

Frequently asked questions

Training from scratch learns all features on the target data alone, while transfer learning starts with useful features already learned from a related source task.