Softmax classifier

In the previous section, we showed how to access and manipulate the MNIST dataset. In this section, we will see how to address the classification problem of handwritten digits via the TensorFlow library.

We'll apply the concepts taught to build more models of neural networks in order to assess and compare the results of the different approaches followed. The first feed-forward network architecture that will be implemented is represented in the following figure:

The softmax neural network architecture

The hidden layer (or softmax layer) of the network consists of 10 neurons, with a softmax transfer function. Remember that it is defined so that its activation is a set of positive values with total sum equal to 1; this means that the j^th value of the output is the probability that j is the class that corresponds with the network input.

Let's see how to implement our neural network model...