I am reading about autoencoder these days. Autoencoders actually reconstruct the original input and also help in dimensionality reduction as below :
At encoder:
X : the input data
c: the resulted code after encoder.
At decoder:
c: The input data
X' : is the output where (X - X') should be almost equals to 0.
My question is about the input X and c. What should it be ? can X be binary data or decimal ? .. the same of C, can it be binary? or it must be decimal.
The other thing, encoder and decoder must be together or we can use them separately? I mean we use the encoder to get the output of encoder c, and then we can modify or change c then we use the decoder, Is that possible? What's the percentage of c compared with X, .. I mean the size of c should be half, quarter the size of X ? or it can be whatever ...
Thank you.