Event
Zwei Netze lernen gegeneinander
In 2014 Ian Goodfellow and colleagues described generative adversarial networks, or GANs. A generator creates examples; a second model, the discriminator, attempts to distinguish training data from generated examples. The generator learns to adjust its outputs using feedback from the discriminator. Original paper
The researchers demonstrated experiments with image datasets. Mathematical results about an ideal equilibrium depend on assumptions and do not mean practical training is always stable or images reliably look real. The preprint appeared in June, followed by the conference paper later in 2014. Original paper
Sources
- Konferenzbeitrag und Vorveröffentlichung: Architektur, Experimente und Grenzen.
Offen
Zeitgenössische Reaktionen und spätere Anwendungen anhand unabhängiger Quellen ergänzen. / Add contemporary responses and later applications using independent sources.