gans-no-guaranteed-nash-equilibrium

IN premise — entries/2026/06/21/wiki-Generative_adversarial_network-chunk-6.md

Created 2026-06-21T09:55:50+00:00

Farnia & Ozdaglar (ICML 2020) proved that GANs do not always have Nash equilibria, establishing a theoretical limitation of adversarial training

Dependents

These beliefs depend on this one:

• IN gan-game-theoretic-foundations-fragile-beyond-original — GAN game-theoretic foundations are fragile beyond the original formulation — equilibrium equivalence (minimax, maximin, Nash) holds only for the original game and not its variants, Nash equilibria are not guaranteed to exist in general (Farnia & Ozdaglar 2020), and the two dominant failure modes (mode collapse and vanishing gradients) represent opposed destabilizing forces that the equilibrium theory does not resolve.
• IN gan-nash-impossibility-mathematically-necessitates-model-level-crisis — Farnia & Ozdaglar's proof that GANs lack guaranteed Nash equilibria provides formal grounding for the GAN-level pragmatism-crisis dynamic — the training instability that exemplifies the field-wide crisis pattern at the model level is not merely an empirical tendency but has a provable theoretical basis in adversarial game structure, strengthening the case that the pragmatism-crisis pattern at the model level is structurally rooted rather than incidental.
• OUT gan-training-stabilizable-given-complete-theory — GAN training would be reliably stabilizable given the complete theoretical characterization (optimal discriminator, JSD minimization, unique equilibrium) and multiple complementary stabilization interventions (non-saturating loss, TTUR, architecture choices) — unless the game-theoretic foundations themselves are limited, with Nash equilibria not guaranteed in general and equilibrium equivalence holding only for the original formulation.