隐藏的层在哪里？

Question

我对自动编码器有点陌生。我有来自Keras（https://blog.keras.io/building-autoencoders-in-keras.html）的代码。我想知道我在这里的代码中的注释正确吗？

input_img = keras.Input(shape=(784,)) # input
encoded = layers.Dense(128, activation='relu')(input_img) # is it hidden layer???
encoded = layers.Dense(64, activation='relu')(encoded) # is it hidden layer???
encoded = layers.Dense(32, activation='relu')(encoded) # is it hidden layer???

decoded = layers.Dense(64, activation='relu')(encoded) # is it hidden layer???
decoded = layers.Dense(128, activation='relu')(decoded) # is it hidden layer???
decoded = layers.Dense(784, activation='sigmoid')(decoded) # output

如果可能的话，你们能解释更多吗？谢谢！

Answer 1

隐藏层是位于输入和输出层（ref）之间的任何层。因此，所有这些都是您网络中的隐藏层：

encoded = layers.Dense(128, activation='relu')(input_img)
encoded = layers.Dense(64, activation='relu')(encoded)
encoded = layers.Dense(32, activation='relu')(encoded)
decoded = layers.Dense(64, activation='relu')(encoded)
decoded = layers.Dense(128, activation='relu')(decoded)

在自动编码器中，有一个特别有趣的隐藏层：网络中的“瓶颈”隐藏层，该层强制对原始输入进行压缩的知识表示。在您的示例中，压缩率为784到32，瓶颈隐藏层为：

encoded = layers.Dense(32, activation='relu')(encoded)

source