Image classification using the MXNet library

The MXNet package was introduced in Chapter 1, Getting Started with Deep Learning, so go back to that chapter for instructions on how to install the package if you have not already done so. We will demonstrate how to get almost 100% accuracy on a classification task for image data. We will use the MNIST dataset that we introduced in Chapter 2, Image Classification Using Convolutional Neural Networks. This dataset contains images of handwritten digits (0-9), and all images are of size 28 x 28. It is the Hello World! equivalent in deep learning. There’s a long-term competition on Kaggle that uses this dataset. The script Chapter5/explore.Rmd is an R markdown file that explores this dataset.

1. First, we will check if the data has already been downloaded, and if it has not, we will download it. If the data is not available at this...