Introduction
VQ-VAE learns a discrete codebook that compresses high-dimensional inputs into combinable representations. DALL·E, VQGAN and several autoregressive image models subsequently draw on the idea of visual tokens. The node concerns the representation of images in units that can participate in learned sequences and generative processes.
View full image ↗Page 4 of the linked paper PDF, preserving its original figures and language. The source file identifies the edition; later revision or reprint dates do not replace the node’s historical date.
Aaron van den Oord; Oriol Vinyals; Koray Kavukcuoglu · original paper · Image rights remain with their respective holders; no open licence is asserted
Source · Neural Discrete Representation Learning ↗RESEARCH ACCOUNT
Representing perceptual material with discrete codes
ANTI-AI ARCHIVE · Revised 2026-10-02
Method and contribution
VQ-VAE uses discrete encoder outputs and a learned prior, in contrast to continuous-latent VAEs. The paper addresses latent variables being ignored by powerful autoregressive decoders and reports applications across images, video and speech. Its codes are learned representations, not predefined artistic-style labels.
Section sources: Neural Discrete Representation Learning
Art-historical interpretation
Editorial interpretation: discrete representation provides context for later visual tokens and staged generation. A codebook is not a neutral taxonomy of visual culture. Training material and representational choices affect which differences remain visible and which become compressed together.
Section sources: Neural Discrete Representation Learning
Sources for this account
Author or conference abstract and version record read; full proofs and experiments were not independently reproduced.
This account expands the archive within the cited scope. Historical interpretation is not presented as the original author’s statement or as a proven causal influence. The account and translation are AI-assisted, pending independent human review. Section references identify evidence without claiming independent verification of every historical statement.
Continue with a comparative question
Compare the methods, forms of control or presentation contexts of this entry and Generating Diverse High-Fidelity Images with VQ-VAE-2.
Compare the methods, forms of control or presentation contexts of this entry and Taming Transformers / VQGAN.
Dates & version record
Date displayed for this node: 2017 · Historical date recorded for the source: 2017
These dates refer to the historical event or recorded version, not this page’s publication date. The original date precision and unresolved questions are retained.
Original sources & further reading
These links lead to the cited paper, article, institution or conference page. External texts retain their source languages.
Neural Discrete Representation Learning
https://arxiv.org/abs/1711.00937
Current cited URL
Provenance, translation & verification
Timeline entry #093 · Historical node
Research materials & supplement references · 1
AI Art Genealogy Research Master · p. 18 · 058
Date as recorded: 2017
Neural Discrete Representation Learning ↗
Neural Discrete Representation Learning
Archive account based on the listed research materials, not a full translation of the linked work. AI-assisted translation; independent human review pending.
The master’s research claims and dates await independent verification.
Cite this node
ANTI-AI ARCHIVE. Neural Discrete Representation Learning. Art-history node #093. https://salondesrefuses.cn/en/art-history/058
For specific historical claims, also cite the original sources above and include your access date. This account is not a full translation of the linked work.
Adjacent nodes follow chronological order; adjacency does not establish direct influence or causation.