We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models

Abstract: Deep generative models are a class of techniques that train deep neural networks to model the distribution of training samples. Research has fragmented into various interconnected approaches, each of which make trade-offs including run-time, diversity, and architectural restrictions. In particular, this compendium covers energy-based models, variational autoencoders, generative adversarial networks, autoregressive models, normalizing flows, in addition to numerous hybrid approaches. These techniques are compared and contrasted, explaining the premises behind each and how they are interrelated, while reviewing current state-of-the-art advances and implementations.
Comments: 20 pages, 9 figures, will appear in IEEE Transactions on Pattern Analysis and Machine Intelligence
Subjects: Machine Learning (cs.LG); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)
MSC classes: 68T01 (Primary), 68T07 (Secondary)
ACM classes: I.5.0; I.4.0; G.3
DOI: 10.1109/TPAMI.2021.3116668
Cite as: arXiv:2103.04922 [cs.LG]
  (or arXiv:2103.04922v4 [cs.LG] for this version)

Submission history

From: Sam Bond-Taylor [view email]
[v1] Mon, 8 Mar 2021 17:34:03 GMT (3940kb,D)
[v2] Wed, 14 Apr 2021 12:50:13 GMT (5105kb,D)
[v3] Fri, 4 Feb 2022 18:57:06 GMT (1856kb,D)
[v4] Mon, 28 Mar 2022 13:32:41 GMT (1856kb,D)

Link back to: arXiv, form interface, contact.