{"id":9531,"date":"2026-07-27T12:30:28","date_gmt":"2026-07-27T10:30:28","guid":{"rendered":"https:\/\/www.kubicek.ai\/?post_type=lexicon&#038;p=9531"},"modified":"2026-07-27T13:19:02","modified_gmt":"2026-07-27T11:19:02","slug":"gan-generative-adversarial-network","status":"publish","type":"lexicon","link":"https:\/\/www.kubicek.ai\/en\/lexicon\/gan-generative-adversarial-network\/","title":{"rendered":"GAN (Generative Adversarial Network)"},"content":{"rendered":"<p class=\"wp-block-paragraph\"><strong>GAN (Generative Adversarial Network)<\/strong> is a generative architecture built on the rivalry of two networks trained simultaneously. The generator receives random noise and tries to turn it into a sample indistinguishable from real data. The discriminator is a binary classifier deciding whether a presented sample comes from the training set or from the generator. The two optimise against each other in a minimax game: the generator minimises what the discriminator maximises. Training therefore amounts to searching for a Nash equilibrium rather than minimising a single function, which makes it notoriously unstable. Typical pathologies are mode collapse, where the generator produces only a handful of variants that reliably fool its opponent, and unbalanced learning, where one network outpaces the other and stops providing a useful signal. Remedies came as architectural and loss modifications such as DCGAN, Wasserstein GAN or the progressive growing of resolution in StyleGAN, which reached photorealistic portraits at the end of the last decade. Since 2022, diffusion models have displaced GANs from most image-generation deployments, as they train more stably and scale better.<\/p>\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n<p class=\"wp-block-paragraph\">Picture an art forger and a gallery curator. The forger paints an imitation and slips it to the curator among the genuine works. The curator guesses which one is fake \u2013 and always learns the outcome. The forger learns from every exposure and paints better; the curator learns from every miss and starts noticing details he used to overlook. After thousands of rounds the forger is so good that the curator is reduced to a coin flip. The danger of this game is that the forger may find one single motif the curator has no eye for, and then keep painting the very same picture forever.<\/p>\n","protected":false},"featured_media":0,"template":"","class_list":["post-9531","lexicon","type-lexicon","status-publish","hentry"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon\/9531","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/lexicon"}],"about":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/types\/lexicon"}],"wp:attachment":[{"href":"https:\/\/www.kubicek.ai\/en\/wp-json\/wp\/v2\/media?parent=9531"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}