Chapter 6: Improving Your Neural Network
In Chapter 4, Deep Learning with Neural Networks, we designed a network that is able to achieve almost 93% accuracy in the training dataset, but that translated to less than 66% accuracy in the validation dataset.
In this chapter, we will continue working on that neural network, with the aim to improve the validation accuracy significantly. Our goal is to reach at least 80% validation accuracy. We will apply some of the knowledge acquired in Chapter 5, Deep Learning Workflow, and we will also learn new techniques that will help us very much, such as batch normalization.
We will cover the following topics:
- Reducing the number of parameters
- Increasing the size of the network and the number of layers
- Understanding batch normalization
- Improving validation with early stopping
- Virtually increasing the dataset size with data augmentation
- Improving validation accuracy with dropout
- Improving validation accuracy with...
