Can we predict from the hidden layer of the neural network?

A new original autoencoder that allows hiding labels inside the hidden layer for later prediction has been developed

“Predicting from the Hidden Layer of a Neural Network!!!???” It's weird, isn't it? Actually this is a very interesting idea; we have developed a new original autoencoder that allows hiding labels inside the hidden layer to predict from it later, rather than the output layer in a completely unsupervised approximation. The idea applies to both numerical and categorical outputs (regression and classification). The question is ……. Why fro the hidden layer?
Check out our recently published work and find out the answer.
Link1: https://www.researchgate.net/publication/350793411_Leveraging_Label_Information_in_a_Knowledge-Driven_Approach_for_Rolling-Element_Bearings_Remaining_Useful_Life_Prediction
Link2:
https://www.mdpi.com/1996-1073/14/8/2163

Note: This version is only the basic version of the one designed in the paper.