Summary

In this chapter, we used deep learning for image classification. We discussed the different layer types that are used in image classification: convolutional layers, pooling layers, dropout, dense layers, and the softmax activation function. We saw an R-Shiny application that shows how convolutional layers perform feature engineering on image data.

We used the MXNet deep learning library in R to create a base deep learning model which got 97.1% accuracy. We then developed a CNN deep learning model based on the LeNet architecture, which achieved over 98.3% accuracy on test data. We also used a slightly harder dataset (Fashion MNIST) and created a new model that achieved over 91% accuracy. This accuracy score was better than all of the other scores that used non-deep learning algorithms. In the next chapter, we will build on what we have covered and show you how we can take advantage of pre-trained models for classification and as building blocks for new deep learning models.

In the next...