Deep Learning (DL) · Beginner level

Beginner level · Module 4: Transfer Learning with Pre-trained Models · Lesson 3 of 3

Fine-Tuning

Unfreeze some pre-trained layers and train them gently to adapt the model further to your data.

In this lesson you will

  • Fine-tune selected layers with a small learning rate
  • Compare fine-tuning with feature extraction

Key concepts

unfreezingsmall learning ratecatastrophic forgettingcomparison

The lesson

Fine-tuning unfreezes some of the later pre-trained layers and trains them together with the new head, adapting the features to your images.

Use a small learning rate for pre-trained layers so they adjust without losing what they learned — large updates can destroy useful features.

Compare with the feature-extraction baseline on validation data. Fine-tuning usually helps when your images differ from the original training images, such as close-up leaf photos.

Step by step

  1. Unfreeze the last block of the network.
  2. Set a smaller learning rate for it.
  3. Fine-tune and compare with the baseline.
  4. Evaluate the better model once on the test set.

Practice activity

Fine-tune the leaf disease classifier and complete the level project evaluation.

Expected result: A fine-tuned model compared with the baseline and evaluated on the test set.

Check your understanding

  1. Why use a small learning rate when fine-tuning?

    Show answer

    To adapt pre-trained features without destroying them.

  2. When does fine-tuning help most?

    Show answer

    When the new images differ from the original training images.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.