van den Oord et al.: VQ-VAE learned discrete latent representations

What changed

Combined learned vector-quantized latent codes, encoder/decoder training and an autoregressive prior to learn discrete representations for generative modeling.

Scientific scope

Combined learned vector-quantized latent codes, encoder/decoder training and an autoregressive prior to learn discrete representations for generative modeling. Vector quantization and variational autoencoding are inherited; the neural discrete-latent training formulation is acquired. Avoiding posterior collapse in examples does not guarantee universal useful codes or global convergence.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Vector quantization and variational autoencoding are inherited; the neural discrete-latent training formulation is acquired. Avoiding posterior collapse in examples does not guarantee universal useful codes or global convergence.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Neural Discrete Representation Learning

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction