A NOVEL INFORMATION-THEORETIC EXPLANATION OF GENERATIVE ADVERSARIAL NETWORKS

Shen, Haochen

Permalink

https://hdl.handle.net/2142/124875 Copy

Description

Title

A NOVEL INFORMATION-THEORETIC EXPLANATION OF GENERATIVE ADVERSARIAL NETWORKS

Author(s)

Shen, Haochen

Issue Date

2022-05-01

Keyword(s)

Generative adversarial networks; channel capacity; unsupervised learning

Date of Ingest

2024-10-15T10:27:24-05:00

Abstract

Generative Adversarial Network (GAN) is widely used for instance generation in multiple fields. It models a zero-sum game between a generator network and a discriminator network while using the Jensen-Shannon divergence between the target distribution and the generated one as the optimization goal. We propose an alternative explanation of such a framework using a binary-input channel. We also show that our proposed goal is equivalent to the original Jensen-Shannon divergence while enforcing a tighter performance bound. We then introduce an additional trainable parameter π with only few modification on the vanilla implementation of GAN. We further experiment with our proposed variant of GAN on MNIST and CIFAR-10, producing ∼ 10 less on the Fr´ echet inception distance and ∼ 0.5 more on the inception score.

Type of Resource

text

Genre of Resource

dissertation/thesis

Language

eng

Owning Collections