Exploiting the Hidden Tasks of GANs: Making Implicit Subproblems Explicit

TL;DR

This paper reveals that GAN training involves two implicit subproblems, and making these explicit leads to improved training methods and results.

Contribution

It introduces a new perspective on GAN training by explicitly decomposing the process into subproblems, enabling better understanding and performance.

Findings

Explicitly modeling subproblems improves GAN training stability.

Using inverse examples as targets enhances generator updates.

Experimental validation shows significant performance gains.

Abstract

We present an alternative perspective on the training of generative adversarial networks (GANs), showing that the training step for a GAN generator decomposes into two implicit subproblems. In the first, the discriminator provides new target data to the generator in the form of "inverse examples" produced by approximately inverting classifier labels. In the second, these examples are used as targets to update the generator via least-squares regression, regardless of the main loss specified to train the network. We experimentally validate our main theoretical result and demonstrate significant improvements over standard GAN training made possible by making these subproblems explicit.