Lossy Compression Text Autoencoder

A text autoencoder combining lossy compression with semantic representation learning. It employs a residual low-dimensional discrete bottleneck architecture to perform sequence compression along the temporal axis, while preserving the semantic information required for downstream tasks.