Bottleneck Layer

The bottleneck layer prevents an autoencoder from trivially copying data by forcing it to preserve only essential information. Because the representation space is narrow, the model is pushed toward more meaningful and compressed latent structure. Without it, an autoencoder may drift toward a weak identity mapping from input to output. It plays a central role in disciplining representation learning.