Back to glossary

Transfer Learning

A technique where a model trained on one task is repurposed as the starting point for a different but related task, dramatically reducing the data and compute needed for the new task.

Transfer learning is why modern AI is accessible to teams without massive datasets or GPU clusters. Instead of training from scratch, you start with a pre-trained model that already understands general patterns (language structure, visual features, audio waveforms) and adapt it to your specific domain. Fine-tuning an LLM on your customer support data is transfer learning in action.

The approach works because lower layers of neural networks learn universal features that transfer across tasks. A vision model's edge detectors are useful whether you are classifying cats or detecting manufacturing defects. Similarly, an LLM's understanding of grammar, logic, and world knowledge transfers to any downstream text task. You only need to adjust the higher layers that map these general representations to your specific output.

Transfer learning has democratized AI by reducing the barrier to entry. Training GPT-4 from scratch costs tens of millions of dollars, but fine-tuning it for your use case might cost a few hundred. For growth teams, this means you can build competitive AI features with modest datasets (hundreds to low thousands of examples) by leveraging the knowledge already encoded in foundation models.

Related Terms