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

We introduce a new distributed policy gradient algorithm and show that itoutperforms existing reward-aware training procedures such as REINFORCE,minimum risk training (MRT) and proximal policy optimization (PPO) in terms oftraining stability and generalization performance when optimizing machinetranslation models. Our algorithm, which we call MAD (on account of using themean absolute deviation in the importance weighting calculation), hasdistributed data generators sampling multiple candidates per source sentence onworker nodes, while a central learner updates the policy. MAD depends cruciallyon two variance reduction strategies: (1) a conditional reward normalizationmethod that ensures each source sentence has both positive and negative rewardtranslation examples and (2) a new robust importance weighting scheme that actsas a conditional entropy regularizer. Experiments on a variety of translationtasks show that policies learned using the MAD algorithm perform very well whenusing both greedy decoding and beam search, and that the learned policies aresensitive to the specific reward used during training.

MAD for Robust Reinforcement Learning in Machine Translation