Adversarial Feature Learning

TL;DR

This paper introduces Bidirectional Generative Adversarial Networks (BiGANs), enabling inverse mapping from data to latent space, and demonstrates their effectiveness for feature learning in supervised tasks.

Contribution

The paper proposes BiGANs, a novel extension of GANs, to learn inverse mappings and improve feature representations for auxiliary supervised tasks.

Findings

BiGANs successfully learn inverse mappings from data to latent space.

Features learned by BiGANs are competitive with other unsupervised methods.

BiGANs enhance feature utility for supervised discrimination tasks.

Abstract

The ability of the Generative Adversarial Networks (GANs) framework to learn generative models mapping from simple latent distributions to arbitrarily complex data distributions has been demonstrated empirically, with compelling results showing that the latent space of such generators captures semantic variation in the data distribution. Intuitively, models trained to predict these semantic latent representations given data may serve as useful feature representations for auxiliary problems where semantics are relevant. However, in their existing form, GANs have no means of learning the inverse mapping -- projecting data back into the latent space. We propose Bidirectional Generative Adversarial Networks (BiGANs) as a means of learning this inverse mapping, and demonstrate that the resulting learned feature representation is useful for auxiliary supervised discrimination tasks, competitive with contemporary approaches to unsupervised and self-supervised feature learning.