MNIST CNN
Learn how to create an MNIST Convolutional Neural Network.
We'll cover the following...
To build our familiarity with convolutions in a neural network, let’s make an MNIST classifier before we try making a GAN that uses them.
We start by making a copy of our previous MNIST classifier given here.
📝 We’ll only need to change the definition of the classifier neural network. The rest of the code for loading the data, viewing images, training the network, and checking classification performance shouldn’t need to change much.
That neural network had an input layer of 784 nodes, fully connected to a middle layer of 200 nodes, which itself was fully connected to an output layer of 10 nodes. The middle layer had a
LeakyReLUactivation and a layer normalization applied after it. The output layer simply had a sigmoid activation applied. That network achieved a really good 97% accuracy on the MNIST test data. ...