Transfer learning (TL)
Transfer learning (TL) is a machine learning approach where a model trained on one task is reused as a starting point for a different but related task. Instead of building a new model from scratch, transfer learning applies knowledge already learned to improve learning speed and performance on the new task.
This approach is especially useful in deep learning, where training models from the ground up often requires large datasets, significant computing power, and long training times.
Transfer learning offers several benefits.
- It reduces training time and computational cost.
- It lowers the need for large labeled datasets,
- It often improves generalization by combining knowledge from multiple data sources.
Because the model has already learned general features, it is less likely to overfit and can perform better in real-world conditions.
However, transfer learning works best when the original and new tasks are reasonably similar. If the tasks or data distributions are too different, performance can suffer—a problem known as negative transfer. For example, a model trained on general image recognition tasks can be adapted to detect medical images like X-rays, but using a language model trained on text to solve image tasks would likely lead to poor results.
