VAEs can be recast as individual neural layers and trained without back-propagation via a multimodal ELBO, yet the resulting shallow classifiers reach only modest accuracy on standard image benchmarks.
Improved Forward-Forward Contrastive Learning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Variational Autoencoder Layer
VAEs can be recast as individual neural layers and trained without back-propagation via a multimodal ELBO, yet the resulting shallow classifiers reach only modest accuracy on standard image benchmarks.