Skip to main content
Contact

AI glossary Models & Architecture

What is transfer learning?

← All glossary terms
In short

Reusing a model trained on one task or dataset as the starting point for a new task, so that it needs far less data and compute.

Explained

Transfer learning is the technique of reusing what a model learnt on one task or domain to solve another. Instead of training from scratch, you start from a pre-trained model, which already captures general patterns of language, images or other data, and adapt it to the new task with a smaller dataset. In language, work such as ULMFiT and BERT made this the standard approach, and today's large language models take it further: a single pre-trained model transfers to new tasks through fine-tuning or simply through prompts. This is why companies can build useful AI without huge task-specific datasets.

How it works

  • Pre-training: a model learns general patterns from a large, broad dataset, such as text from many sources or millions of images.
  • Adaptation: the pre-trained model is adjusted for a specific task with a smaller set of labelled examples.
  • Options for adapting: retrain all the parameters, retrain only the last layers, train small add-on modules, or skip training and use prompts with examples.
  • Evaluation: the adapted model is tested on the target task, because transfer works best when the new task is related to what the model learnt.

Transfer learning, fine-tuning and prompting

Transfer learning is the general idea; fine-tuning is one way to do it, by continuing training on task data. With large language models, transfer often happens without any training: the model applies what it learnt to a new task from instructions and a few examples in the prompt, known as few-shot or in-context learning. Which option to use depends on the amount of data, the required quality and the cost.

When it works and when it does not

Transfer works well when the new task shares patterns with the pre-training data, such as classifying customer emails with a model trained on general text. It works less well when the domain is very different, the vocabulary is highly specialised or the data has a different structure. In those cases, more domain data, continued pre-training or a different base model may be needed.

Frequently asked questions

What is transfer learning in simple terms?

It is reusing a model that has already learnt something general and adapting it to a new, related task. Like a person who already speaks Spanish learning Italian faster, the model needs much less data to learn the new task.

What is the difference between transfer learning and fine-tuning?

Transfer learning is the broad idea of reusing a pre-trained model. Fine-tuning is a common way to do it, by training the model further on data for the new task. Prompting a large language model with examples is another form of transfer.

Why is transfer learning important?

It cuts the data, time and compute needed to build a useful model. Most enterprise AI today starts from a pre-trained model rather than training one from scratch.

What are examples of transfer learning?

Adapting an image model trained on general photos to detect defects on a production line, or adapting a language model to classify support tickets or extract data from contracts. Large language models answering new tasks from a prompt are also an example.

Is a large language model an example of transfer learning?

Yes. LLMs are pre-trained on large amounts of text and then applied to many tasks they were not explicitly trained for, through instructions, examples or fine-tuning. That reuse of general knowledge is transfer learning.

Where it shows up at Thinkia

The Mesh

Further reading

Sources

Updated: 2026-10-21

Want to see how this applies to your organisation?

Talk to an AI Expert