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

Variational autoencoders (VAEs) are one of the powerful unsupervised learningframeworks in NLP for latent representation learning and latent-directedgeneration. The classic optimization goal of VAEs is to maximize the EvidenceLower Bound (ELBo), which consists of a conditional likelihood for generationand a negative Kullback-Leibler (KL) divergence for regularization. Inpractice, optimizing ELBo often leads the posterior distribution of all samplesconverge to the same degenerated local optimum, namely posterior collapse or KLvanishing. There are effective ways proposed to prevent posterior collapse inVAEs, but we observe that they in essence make trade-offs between posteriorcollapse and hole problem, i.e., mismatch between the aggregated posteriordistribution and the prior distribution. To this end, we introduce new trainingobjectives to tackle both two problems through a novel regularization based onthe probabilistic density gap between the aggregated posterior distribution andthe prior distribution. Through experiments on language modeling, latent spacevisualization and interpolation, we show that our proposed method can solveboth problems effectively and thus outperforms the existing methods inlatent-directed generation. To the best of our knowledge, we are the first tojointly solve the hole problem and the posterior collapse.

Improving Variational Autoencoders with Density Gap-based Regularization