English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Autoregressive sequence models achieve state-of-the-art performance indomains like machine translation. However, due to the autoregressivefactorization nature, these models suffer from heavy latency during inference.Recently, non-autoregressive sequence models were proposed to reduce theinference time. However, these models assume that the decoding process of eachtoken is conditionally independent of others. Such a generation processsometimes makes the output sentence inconsistent, and thus the learnednon-autoregressive models could only achieve inferior accuracy compared totheir autoregressive counterparts. To improve then decoding consistency andreduce the inference cost at the same time, we propose to incorporate astructured inference module into the non-autoregressive models. Specifically,we design an efficient approximation for Conditional Random Fields (CRF) fornon-autoregressive sequence models, and further propose a dynamic transitiontechnique to model positional contexts in the CRF. Experiments in machinetranslation show that while increasing little latency (8~14ms), our model couldachieve significantly better translation performance than previousnon-autoregressive models on different translation datasets. In particular, forthe WMT14 En-De dataset, our model obtains a BLEU score of 26.80, which largelyoutperforms the previous non-autoregressive baselines and is only 0.61 lower inBLEU than purely autoregressive models.

Fast Structured Decoding for Sequence Models