TensorFlow needs to be installed and loaded in the environment:

require(tensorflow)

Dependencies need to be loaded:

require(imager)
require(caret)

The MNIST dataset needs to be loaded. The dataset is normalized using the following script:

# Normalize Dataset
normalizeObj<-preProcess(trainData, method="range")
trainData<-predict(normalizeObj, trainData)
validData<-predict(normalizeObj, validData)