Hands-On Artificial Intelligence for Beginners

Forward propagation

Forward propagation is the process of how information flows through our network during the learning phase of training. Before we walk through the process, let's return to our MNIST example. First, we need to initialize the feedforward network that we precedingly created. To do that, we can simply create an instance of the function, and feed it the input placeholder, as well as the weight and bias dictionaries:

network_output = feedforward_network(x, weights, biases)

This model instance gets fed into our loss function that we preceedingly defined: 

loss_func = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=network_output, labels=y))

Which subsequently gets fed into the optimizer we defined: 

training_procedure ...