Hands-On Deep Learning for IoT

Creating an AE classifier

Next, we will re-train the auto_encoder model by making the first three layers trainable as True instead of keeping them as False:

for layer in auto_encoder.layers[0:3]:    layer.trainable = True  

Alternatively, we can pop off the first three layers as follows:

for i in range(number_of_layers_to_remove):    auto_encoder.pop()

Then, we add fully connected layers in front, with the BatchNormalization layer is followed by the first dense layer. Then, we add another dense layer, followed by the BatchNormalization and Dropout layers. Then, we place another dense layer, followed by a GaussionNoise layer and a Dropout layer, before we finally have the softmax layer:

auto_encoder.add(Dense(128, input_dim=64, activation='relu', ...