Learn AI · Level 2 · How AI works · step 12 of 18 · word 3 of 6
What is transfer learning?
Transfer learning means starting from a model that already learned one task, often from huge amounts of data, and adapting it to a new task with far less data.
In plain English
Someone who already plays the piano learns the organ faster than a beginner. In the same way, a model that has learned to recognise thousands of objects already knows what edges, fur and wheels look like. You only need to teach it the last step.
An example
Google’s official TensorFlow tutorial takes MobileNet, a model trained on 1.4 million ImageNet photos, and adapts it to tell cats from dogs using just 2,000 training pictures. Training from scratch would need far more data and computing power.
Watch out
Transfer learning is why most teams never train a model from zero. Today’s foundation models take the idea furthest: one big model, adapted to many jobs with fine-tuning or LoRA. It works best when the new task resembles what the model learned first.