AI & Animation Education

Home  /  Knowledge Base  /  Generative Adversarial Networks (GAN)

Research · Technical

Generative Adversarial Networks (GAN)

NeurIPS 2014 · Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y. · Jun 2014

Key points

  1. GANs pit a generator against a discriminator in a minimax game, learning to synthesise realistic images without an explicit density model.
  2. The framework is the named origin of StyleGAN, pix2pix, CycleGAN and the whole pre-diffusion generative-image lineage.
  3. It is universal baseline literacy in AI-art and creative-AI courses.

Summary

Goodfellow et al. (2014) introduced Generative Adversarial Networks, a training framework in which two neural networks compete: a generator attempts to produce realistic samples and a discriminator attempts to distinguish them from real data. The adversarial dynamic drives both networks to improve until the generator produces outputs indistinguishable from training data. The GAN framework is the named origin of StyleGAN, pix2pix and CycleGAN, making it prerequisite knowledge for any course covering generative image synthesis. Understanding GANs is foundational for animation educators introducing AI image generation, as it underpins the entire pre-diffusion era of generative tools used in character, concept and visual-development workflows.

Source

Source: NeurIPS 2014 ↗ (Research)

Cite this item

Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A. & Bengio, Y. (2014). ‘Generative Adversarial Networks (GAN)’, NeurIPS 2014. Available at: https://arxiv.org/abs/1406.2661

Download .risDownload .bib

Your reference manager can also read this page directly: with the Zotero (or Mendeley) browser connector installed, save it straight to your library. Whole-collection exports: RIS, BibTeX, CSL-JSON.