Transfer Learning

Definition

The technique of leveraging knowledge gained from training on one task to enhance performance on a different but related task, improving efficiency.

Use Cases

Provider Equivalents

Frequently Asked Questions

What's the difference between transfer learning and fine-tuning?
Transfer learning is the broader idea: reuse knowledge from a model trained on one task to help with another related task. Fine-tuning is a common way to do transfer learning: you take a pre-trained model and continue training it on your new dataset (often with a smaller learning rate) so it adapts to your specific task.
When should I use transfer learning?
Use transfer learning when (1) you have limited labeled data, (2) your task is similar to a well-known domain with strong pre-trained models (images, text, speech), (3) you need faster training and iteration, or (4) training from scratch would be too expensive. Avoid it when your data is very different from the pre-training domain (e.g., medical imaging vs. everyday photos) unless you can validate performance carefully.
How much does transfer learning cost?
Costs depend on compute time (GPU/TPU hours), model size, training method (full fine-tune vs. training only the head vs. parameter-efficient tuning), dataset size, and experimentation cycles. Transfer learning is often cheaper than training from scratch because you train fewer steps and sometimes fewer parameters, but large foundation models can still be expensive to fine-tune. Also budget for data labeling, storage, and deployment inference costs.

Category: ai-ml

Difficulty: advanced

Related Terms

See Also