Equivalence of GANs and Zero Sum Games

Summary

  • A game theoretic analysis of Generative Adversarial Networks, exploiting the two-player minimax game modelling
  • Collated research papers to show that the training process is a two-player zero sum game, under certain constraints, satisfying the conditions of MiniMax theorem; allows for Nash equilibrium to be minmax/maxmin value, addressing the issue of instability in GANs

Related