Mini-Max Optimization Design of Generative Adversarial Networks (GAN)

Nested bi-level optimization and equilibrium seeking objective

Introduction

Generative Adversarial Networks (GAN) demonstrated outstanding performance in generating realistic synthetic data which were indistinguishable from the real data. Unfortunately, GAN caught the public’s attention because of its illegit applications, Deep Fake. (Knight, 2018)

As its name suggests, Generative Adversarial Nets (GAN) is composed of two networks: the generative network (the generator) and the adversarial network (the discriminator). Incorporating an adversarial scheme into its architecture makes GAN a special type of …

generative-adversarial generative-networks optimization