Generative Adversial Networks. A 2019 Review

0518921 - ÚI 2020 CZ eng A - Abstract
Coufal, David
Generative Adversial Networks. A 2019 Review.
Proceedings of the 22nd Czech-Japan Seminar on Data Analysis and Decision Making (CJS’19). Praha: MatfyzPress, 2019 - (Inuiguchi, M.; Jiroušek, R.; Kratochvíl, V.). s. 25-27. ISBN 978-80-7378-400-3.
[CJS 2019. Czech-Japan Seminar on Data Analysis and Decision Making /22./. 25.09.2019-28.09.2019, Nový Světnov]
R&D Projects: GA ČR(CZ) GA18-23827S
Institutional support: RVO:67985807
http://cjs.utia.cas.cz/proceedings.pdf

The generative adversial networks (GANs) represent an exciting concept that lies at the borderline between probabilistic modeling and machine learning. Despite the short history (since 2014) the GANs have caused a dramatic shift in several machine learning fields, however, mostly in generating photorealistic pictures of different objects. The basic idea the GANs are based on is conceptually simple, however, concrete implementations are still partly an art rather than established science and are tightly interconnected with programming in deep-learning frameworks such as TensorFlow or PyTorch to mention the most popular ones. The purpose of this contribution is to present a brief review of the GAN implementations milestones that paved the way to their current success. The list is not exhaustive as hundreds of implementations are available in several GANs ZOOs, see e.g., [5], but we try to be representative.
Permanent Link: http://hdl.handle.net/11104/0303927