What Is Transfer Learning? Reusing What a Model Already Knows
Instead of training a new model from scratch for every task, transfer learning reuses a model's existing knowledge as a head start.
Starting from knowledge, not from zero
Transfer learning takes a model already trained on one broad task and adapts it for a different, often narrower one — reusing the general patterns it already learned instead of starting training from randomly initialized parameters. It's the reason fine-tuning an existing model is so much cheaper than training a new one from scratch.
Why it works
A lot of what a large model learns during pretraining — grammar, general world knowledge, reasoning patterns — is broadly useful across many different downstream tasks, not specific to any one of them. Transfer learning is essentially betting that this general knowledge is a better starting point than nothing at all, which in practice it reliably is.
Where it shows up in practice
Every foundation model used for a specific downstream product is an example of transfer learning at work — the base model's broad training gets reused and specialized, rather than any provider training a fresh model for every single application.
Frequently Asked Questions
What is transfer learning?
Reusing a model already trained on one broad task as the starting point for a different, often narrower task — instead of training a new model from scratch.
Why is transfer learning cheaper than training from scratch?
Because it reuses general knowledge (grammar, reasoning, world knowledge) the model already learned, requiring far less additional training data and compute to specialize it further.