Squeezing bottlenecks: exploring the limits of autoencoder semantic representation capabilities

Open in new window