Cooperate or Compete: A New Perspective on Training of Generative Networks

TL;DR

This paper proposes a new perspective on training generative adversarial networks (GANs) by modeling their training as an infinitely repeated game where modules cooperate based on their performance dynamics, leading to improved learning.

Contribution

It introduces a novel training paradigm for GANs that emphasizes cooperation between modules based on performance, contrasting with traditional non-cooperative training methods.

Findings

Modules perform better when modeled as an infinitely repeated game.

Cooperation based on performance dynamics accelerates learning.

The approach offers a new perspective on GAN training stability.

Abstract

GANs have two competing modules: the generator module is trained to generate new examples, and the discriminator module is trained to discriminate real examples from generated examples. The training procedure of GAN is modeled as a finitely repeated simultaneous game. Each module tries to increase its performance at every repetition of the base game (at every batch of training data) in a non-cooperative manner. We observed that each module can perform better and learn faster if training is modeled as an infinitely repeated simultaneous game. At every repetition of the base game (at every batch of training data) the stronger module (whose performance is increased or remains the same compared to the previous batch of training data) cooperates with the weaker module (whose performance is decreased compared to the previous batch of training data) and only the weaker module is allowed to increase its performance.