All terms

Transfer Learning

Reusing knowledge from a model trained on one task or dataset to improve performance on a related task with less data.

Deep Learning1 min read

Definition

Transfer learning pretrains on a large source task (ImageNet, web text), then adapts to a target task via fine-tuning or feature extraction.

It is why modern AI works with limited labeled data — including low-resource languages.

In simple terms

Learning piano makes learning another instrument faster. Transfer learning is that skill carryover for neural networks.

Where you see it

  • BERT/GPT pretrained on web text, fine-tuned for classification.
  • Vision models pretrained on ImageNet for medical imaging.
  • Multilingual models adapted for Somali NLP.

How it works

  1. 1.Pretrain

    Learn general representations on large data.

  2. 2.Adapt

    Fine-tune or freeze backbone and train a new head.

  3. 3.Evaluate

    Confirm gains on the target domain.

Why it matters

  • Transfer learning is the default strategy for practical deep learning today.

Often confused

  • Transfer learning and fine-tuning are unrelated.

    Fine-tuning is the most common way to perform transfer learning.