Adversarial Fisher Vectors for Unsupervised Representation Learning

TL;DR

This paper explores the use of Fisher Vectors derived from GANs viewed as Energy Based Models, demonstrating their effectiveness in unsupervised feature extraction for classification and similarity tasks.

Contribution

It introduces a novel approach to derive Fisher Vectors from GANs in the EBM framework, enabling effective unsupervised feature extraction.

Findings

Fisher Vectors from GANs perform competitively in classification tasks.

The method allows computation of a distance metric between examples.

GAN-induced Fisher Vectors are useful for perceptual similarity.

Abstract

We examine Generative Adversarial Networks (GANs) through the lens of deep Energy Based Models (EBMs), with the goal of exploiting the density model that follows from this formulation. In contrast to a traditional view where the discriminator learns a constant function when reaching convergence, here we show that it can provide useful information for downstream tasks, e.g., feature extraction for classification. To be concrete, in the EBM formulation, the discriminator learns an unnormalized density function (i.e., the negative energy term) that characterizes the data manifold. We propose to evaluate both the generator and the discriminator by deriving corresponding Fisher Score and Fisher Information from the EBM. We show that by assuming that the generated examples form an estimate of the learned density, both the Fisher Information and the normalized Fisher Vectors are easy to compute. We also show that we are able to derive a distance metric between examples and between sets of examples. We conduct experiments showing that the GAN-induced Fisher Vectors demonstrate competitive performance as unsupervised feature extractors for classification and perceptual similarity tasks. Code is available at \url{https://github.com/apple/ml-afv}.