Byte-Sized Deep Learning Series: Understanding Transfer Learning

Description:

Deep learning models are data-hungry; but what if you could leverage a pre-trained network to solve your problem with much less data?

In this 90-minute session, we’ll explore transfer learning, a powerful technique that enables you to use knowledge gained from one task and apply it to another.

You’ll learn about the types of transfer learning (fine-tuning vs feature extraction), the benefits of transfer learning, and how to implement these strategies using Keras and pre-trained models.

We’ll dive into feature extraction, where we use frozen pre-trained layers, and fine-tuning, where we allow some layers to be updated during training.

Through hands-on examples on an image dataset, you’ll see how to apply transfer learning to boost model performance with less data and fewer resources.

Whether you’re tackling a new classification problem or trying to improve an existing model, this session will equip you with the knowledge to use transfer learning effectively in your projects.

Duration: 90 mins

Course Code: BDT497

Learning Objectives:

After this course, you will be able to:

• Introduction to Transfer Learning
• Feature Extraction with Pre-Trained model
• Fine-Tuning pre-trained models

• Learners familiar with Keras/TensorFlow basics and willing to apply pre-trained models to real-world problems

• Machine learning practitioners and deep learning students with a solid understanding of neural networks. Ideal for those looking to improve model performance quickly by leveraging pre-trained networks

Course Outline:

1. Introduction to Transfer Learning
   • What is Transfer Learning? Reusing knowledge from one model for a different task
   • Types of Transfer Learning: Feature Extraction, Fine Tuning
   • Benefits of transfer learning
2. Feature Extraction with Pre-Trained Models
   • The idea: using a pre-trained model as a fixed feature extractor
   • Understand how feature extraction works
   • Freezing weights, removing layer and adding a layer
   • Hands-on: Perform feature extraction when performing image classification
3. Fine-Tuning a Pre-Trained Model
   • Fine tuning: unfreezing a few top layers and allowing them to train
   • Understand when to fine tune and how to fine tune
   • Hands-on: Fine tune a model for image classification

Training material provided: Yes (Digital format)

Hands-on Lab: Instructions will be provided to install Jupyter notebook and other required python libraries. Students can opt to use ‘Google Colaboratory’ if they do not want to install these tools