How to reuse hidden layers from autoenocder for classification task in tensorflow

Question

can someone explain with example that how to reuse hidden layers of autoenocder for classification task from neural networks, I want to use two layers of my autoencoder in my multi-layer perceptron model in tensorflow

Answer 1

• Once you have trained the auto enocder then hidden representation can be used for classification
• After training the autoencoder you freeze the weights of your autoencoder model
• Now you make a forward pass from input layer to the hidden layer whose output is the hidden representation
• The output of the hidden representation can be used as input to any normal classifier like SVM or may be another neural network like MLP
• Now You can use only one layer of your autoencoder for classsification
• If you are using two layer of the autoenocder (Which I have not seen anyone doing so I think its a bad idea) then you have to concatenate the results of both the layer and now the concatenated version will be input to another classifier like SVM
• If you have doubts in coding this then show me your code and i Will tell you in code what are the further steps required