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) combine latent variables with amortizedvariational inference, whose optimization usually converges into a triviallocal optimum termed posterior collapse, especially in text modeling. Bytracking the optimization dynamics, we observe the encoder-decoderincompatibility that leads to poor parameterizations of the data manifold. Weargue that the trivial local optimum may be avoided by improving the encoderand decoder parameterizations since the posterior network is part of atransition map between them. To this end, we propose Coupled-VAE, which couplesa VAE model with a deterministic autoencoder with the same structure andimproves the encoder and decoder parameterizations via encoder weight sharingand decoder signal matching. We apply the proposed Coupled-VAE approach tovarious VAE models with different regularization, posterior family, decoderstructure, and optimization strategy. Experiments on benchmark datasets (i.e.,PTB, Yelp, and Yahoo) show consistently improved results in terms ofprobability estimation and richness of the latent space. We also generalize ourmethod to conditional language modeling and propose Coupled-CVAE, which largelyimproves the diversity of dialogue generation on the Switchboard dataset.

On the Encoder-Decoder Incompatibility in Variational Text Modeling and Beyond