GAN (adversarial)
A generator turns noise into fakes while a discriminator learns real from fake; the loss back-propagates to both networks.
Rendering…
Make it your own.
{
"nodes": [
{ "id":"z", "type":"rounded", "label":"Noise z", "x":40, "y":80, "fill":"#dbeafe" },
{ "id":"gen", "type":"neural-net", "label":"Generator", "x":260, "y":80, "fill":"#ede9fe" },
{ "id":"fake", "type":"rounded", "label":"Fake", "x":500, "y":80, "fill":"#fef3c7" },
{ "id":"real", "type":"cylinder", "label":"Real data", "x":260, "y":320, "fill":"#dcfce7" },
{ "id":"disc", "type":"neural-net", "label":"Discriminator", "x":730, "y":200, "fill":"#ede9fe" },
{ "id":"loss", "type":"donut-3d", "label":"Loss", "x":970, "y":200, "fill":"#fee2e2" }
],
"edges": [
{ "id":"e1","source":"z","target":"gen" },
{ "id":"e2","source":"gen","target":"fake" },
{ "id":"e3","source":"fake","target":"disc" },
{ "id":"e4","source":"real","target":"disc" },
{ "id":"e5","source":"disc","target":"loss","label":"real / fake" },
{ "id":"e6","source":"loss","target":"gen","label":"backprop" }
]
}