Preventing Posterior Collapse with DVAE for Text Modeling
Tianbao Song, Zongyi Huang, Xin Liu, Jingbo Sun • Entropy
(2025)
This paper introduces a novel variational autoencoder model termed DVAE to prevent posterior collapse in text modeling. DVAE employs a dual-path architecture within its decoder: path A and path B. Path A makes the direct input of text insta...
DOAJ
Open Access (DOAJ)
DOI:
10.3390/e27040423