Wait, embeddings were used for classification for a long time now. Can somebody explain what is new here? edit: ah, looked at the paper, they did it unsupervis

Wait, embeddings were used for classification for a long time now. Can somebody explain what is new here? edit: ah, looked at the paper, they did it unsupervised, with a sparse autoencoder.