Neural Network Programming with TensorFlow

Fully connected layers

In the fourth and fifth layers, we define fully connected layers:

layer_fc1 = new_fc_layer(input=layer_flat,                         num_inputs=num_features,                         num_outputs=fc_size,                         use_relu=True)

where

• layer_flat: the last layer flattened
• num_features: number of features
• fc_size: number of outputs

The following image shows the values that are passed to new_fc_layer():

print(layer_fc1)

The value of the print is as follows:

Tensor("Relu_3:0", shape=(?, 128), dtype=float32)

Next is the fully connected layer 2, where the function takes the following parameters:

• layer_fc1: the output from the first fully connected layer
• num_inputs: 128
• num_inputs ...