Gabriele Prato, Mathieu Duchesneau, Sarath Chandar +1
The paper proposes a near‑lossless neural method for encoding entire sentences and their sub‑sequences into feature‑rich representations that can be reconstructed, and evaluates th…