Deep Learning (DL) · Beginner level

Beginner level · Module 4: Transfer Learning with Pre-trained Models · Lesson 1 of 3

Why Transfer Learning Works

Understand how features learned on large data sets can be reused for new tasks.

In this lesson you will

  • Explain transfer learning and when it helps
  • Choose a pre-trained model and check its licence

Key concepts

pre-trained modelgeneral vs specific featuresmodel hublicence

The lesson

Training a large network from scratch needs huge data sets and computing power. A model pre-trained on millions of images has already learned general features — edges, textures, shapes — that are useful for many new tasks.

Transfer learning reuses these features and trains only what is specific to the new task, so a few hundred labelled photos of local crops can be enough.

Pre-trained models are available from libraries such as torchvision and model hubs. Check each model's licence and what data it was trained on, since that shapes its strengths and biases.

Step by step

  1. List three pre-trained image models in torchvision.
  2. Compare their size and accuracy.
  3. Read the licence of one model.
  4. Choose a model for slow laptops and justify it.

Practice activity

Choose a pre-trained model for the leaf disease project and justify the choice.

Expected result: A justified choice covering size, accuracy, licence and training data.

Check your understanding

  1. Why can transfer learning work with small data sets?

    Show answer

    The model already learned general features from large data sets.

  2. Why check a pre-trained model's training data?

    Show answer

    It shapes what the model does well and its possible biases.

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.