PyTorch Pocket Reference

Chapter 4. Neural Network Development Reference Designs

In the previous chapter we covered NN development process at a high level, and you learned how to implement each stage in PyTorch. The examples in that chapter focused on solving an image classification problem with the CIFAR-10 dataset and a simple fully connected network. CIFAR-10 image classification is a good academic example to illustrate the NN development process, but there’s a lot more to developing deep learning models with PyTorch.

This chapter presents some additional reference designs for NN development with PyTorch. Reference designs are code examples that you can use as a reference to solve similar types of problems.

Indeed, the set of reference designs in this chapter merely scratches the surface when it comes to the possibilities of deep learning; however, I’ll attempt to provide you with enough variety to assist you in the development of your own solutions. We will use three examples to process a variety of data, design different model architectures, and explore other approaches to the learning process.

The first example uses PyTorch to perform transfer learning to classify images of bees and ants with a small dataset and a pretrained network. The second example uses PyTorch to perform sentiment analysis using text data to train an NLP model that predicts the positive or negative sentiment of movie reviews. And the third example uses PyTorch to demonstrate generative learning by training a generative adversarial ...