Unrolled Generative Adversarial Networks

TL;DR

This paper proposes an unrolled optimization technique for GANs that stabilizes training, reduces mode collapse, and improves diversity by approximating the optimal discriminator during generator updates.

Contribution

It introduces a novel unrolled optimization method for GAN training, enhancing stability and diversity compared to traditional approaches.

Findings

Reduces mode collapse in GAN training.

Stabilizes training of complex recurrent generators.

Increases data distribution coverage and diversity.

Abstract

We introduce a method to stabilize Generative Adversarial Networks (GANs) by defining the generator objective with respect to an unrolled optimization of the discriminator. This allows training to be adjusted between using the optimal discriminator in the generator's objective, which is ideal but infeasible in practice, and using the current value of the discriminator, which is often unstable and leads to poor solutions. We show how this technique solves the common problem of mode collapse, stabilizes training of GANs with complex recurrent generators, and increases diversity and coverage of the data distribution by the generator.