Tweeted By @harvardnlp

(arxiv) Experiments with Variational Attention (https://t.co/IB51FIdjTt, https://t.co/gwt7uZszAg): fast training for latent var. attention (e.g. hard attention) with accuracy like soft attention. Learns sharp attention distributions (red), and useful posterior corrections (blue) pic.twitter.com/M8rS3B3fqK

— harvardnlp (@harvardnlp) July 11, 2018