0/1 Classification using Autoencoders

Hello Everyone,

Was just pondering this the other day. I have a chain of stacked denoising autoencoders that are taking in a signal and reconstructing it based on the previous signal. Now, would it be possible to take this output (some floating point vector) and do a 0/1 output on it WITHOUT supervised criterion?

I am just trying to evaluate how well the signal is being reconstructed w.r.t the original. I was thinking along the lines of:

output = sigmoid(y'x), where y is the reconstruction of the autoencoder and x is the input signal (' = transpose). 

If someone has any ideas or links to papers I would really appreciate it!