This seems like an old school and specific way of describing a much larger goal using NNs and auto encoders. To take data and reduce its dimensionality, while

This seems like an old school and specific way of describing a much larger goal using NNs and auto encoders. To take data and reduce its dimensionality, while getting a (near) perfect reconstruction of the original input from the latent space data. Reducing (haha pun) this broad effort to just "data compression" leaves a lot of nuance and use out of sight.