Transfer Learning
Core ConceptsUsing knowledge a model learned from one task or dataset to improve performance on a different but related task — the principle behind fine-tuning.
Transfer learning is the idea that AI knowledge is portable. A model trained to understand English text can be adapted to understand medical text. A model trained on millions of images can be fine-tuned to identify specific products. The base knowledge transfers to new domains.
This is why fine-tuning works: rather than training a model from scratch on your specific data (which would require enormous amounts), you start with a pre-trained model that already understands language/images generally, then adapt it to your domain with relatively little data.
Transfer learning revolutionized AI practical applications. Before transfer learning, each new task needed a model trained from scratch. Now, a foundation model like GPT-4 or Llama serves as the base for thousands of specialized applications.
Real-World Example
When a company fine-tunes Llama on their customer support data they're using transfer learning — the model's general language ability transfers to their specific domain.
Related Terms
More in Core Concepts
FAQ
What is Transfer Learning?
Using knowledge a model learned from one task or dataset to improve performance on a different but related task — the principle behind fine-tuning.
How is Transfer Learning used in practice?
When a company fine-tunes Llama on their customer support data they're using transfer learning — the model's general language ability transfers to their specific domain.
What concepts are related to Transfer Learning?
Key related concepts include Fine-tuning, Pre-training, Foundation Model, Few-Shot Learning. Understanding these together gives a more complete picture of how Transfer Learning fits into the AI landscape.