TensorFlow Beginner: Transfer Learning

Name: TensorFlow Beginner: Transfer Learning
Author: Josh Turner

Learn how to improve the performance of your machine learning models using transfer learning and pretrained models

Preview the first lesson free — get full access to all 4 lessons.

Course: On-Demand

Hello, my name is Josh Turner, and these lessons you'll learn about transfer learning. You'll build your very own transfer learning model. You'll train and compile your model and test it on some test data. Finally, you'll learn how to validate and tune your model to get better predictions and results. Let's get started. In this lesson, you'll learn about transfer learning and build your very own transfer learning model. We'll describe transfer learning and then import the necessary libraries and the dataset to build our own transfer learning model. The first thing we should do is describe transfer learning. Transfer learning is the intentional transfer of information or knowledge from one entity to another. In machine learning, that means using the information or knowledge learned from one model to build another model. Let's see how this is done. To start IntensorFlow, we should import the necessary libraries. Let's import TensorFlow as TF. Additionally, from TensorFlow, we need to import the Keros library. And then from Keros, we need to import the layers library. Okay. Go ahead and type and run these lines of code to make sure everything works okay on your local machine. All right, so far so good. The next step for us is to import a dataset. We'll import the fashion MNIST dataset, which is comprised of many images of different types of clothing. To do that, set four variables, ABUNV equal to Keros.datasets.fashion.load.data. In addition, we'll set an input shape variable of 224 to 24,3. This is the input parameter shape for the transfer learning model that we will build in a few minutes. Go ahead and run these lines of code. All right, so far so good. Now that we have our input data set and an input shape, we can start to build our very own transfer learning model. We'll use the mobile net V2 model in TensorFlow. This model has been trained on thousands of different images and hasn't learned lots of information about the features and representations of these images. We'll use the knowledge from this model to build our very own classifier. For now though, we'll build our very own transfer learning model by using the mobile net V2 model. To do this, type the following code. Create a variable called base model. Set base model equal to tf.keros.applications mobile net V2. The mobile net V2 model takes some different parameters. The input shape will be the input shape that we defined earlier. We will not include the top because we will use weights that are already pre-imported from image net. So set your include top equal to false. And for weights and double quotes type image net. Once you've typed these lines of code, go ahead and run them to make sure you have an instance of your very own transfer learning model. The final step in making our very own transfer learning model is to freeze the model layers. That means that the model has been pre-trained on some given weights and frozen so that we can use it for our own purposes. To do that, we simply need to use the trainable attribute for the model and set it equal to false. That means typing base model dot trainable equal to false. Go ahead and run this final line of code and we have created our very own transfer learning model. Thanks for watching. Stay tuned for the next lesson when we will train and compile a transfer learning model.

Beginner Provider : Josh Turner 4 Lessons · 20m in Arabic, German, English, Spanish, French, Portuguese, Chinese

Course Description

Transfer learning is a powerful technique in machine learning that involves using pretrained models to improve the performance of your own models. In these lessons, you will learn about the different benefits of transfer learning and how to use pretrained models to improve the performance of your machine learning projects.

You will start by exploring the different True/False statements related to transfer learning, such as the need to reshape data for pretrained models and the importance of using binary cross-entropy for two-class classification problems. You will also learn about popular image models used in transfer learning such as MobileNetV2, and where to download pretrained TensorFlow models.

Moving forward, you will discover the different preprocessing steps involved in transfer learning, such as reshaping data and normalizing data. You will also explore the benefits of batching, such as speeding up your model and making large datasets easier to process.

You will learn how to use tools such as validation datasets and confusion matrices to fine-tune your model and improve its performance. Additionally, you will explore the different parameters involved in transfer learning, such as the number of epochs for training and the freezing of weights.

What You'll Learn

Understand the concept of transfer learning and its benefits in machine learning
Use pretrained TensorFlow models such as MobileNetV2 to improve your machine learning projects
Preprocess data for pretrained models by reshaping and normalizing it
Build, compile, and train a transfer learning model
Fine-tune your model using validation datasets and confusion matrices
Test and validate your model's fit on examples

Key Takeaways

Transfer learning uses pretrained models to improve the performance of your own machine learning models.
MobileNetV2 is a popular image model used in transfer learning, and pretrained TensorFlow models can be downloaded for use.
Preprocessing for transfer learning includes reshaping data and normalizing data so it works with pretrained models.
Batching speeds up your model and makes large datasets easier to process.
Parameters in transfer learning include the number of epochs for training and the freezing of weights.

Frequently Asked Questions

What will I learn in this course?

You will learn the concept of transfer learning and its benefits, how to use pretrained models like MobileNetV2 to improve your machine learning projects, how to preprocess data by reshaping and normalizing it, and how to fine-tune your model using validation datasets and confusion matrices.

What skills does this course cover?

The course covers Deep Learning, Machine Learning Model Training, Supervised Learning, TensorFlow, Training Datasets, and Transfer Learning.

What topics are covered in the lessons?

The lessons cover building a transfer learning model, compiling and training the model, testing the model on examples, and validating the model fit.

What preprocessing steps does this course teach for transfer learning?

It teaches preprocessing steps such as reshaping data and normalizing data, and explains the benefits of batching, including speeding up your model and making large datasets easier to process.

What model parameters are discussed in this course?

The course explores parameters involved in transfer learning, such as the number of epochs for training and the freezing of weights.