Venn GAN: Discovering Commonalities and Particularities of Multiple Distributions

TL;DR

Venn GAN introduces a novel generative adversarial network architecture that models multiple data distributions simultaneously, effectively capturing their shared features and unique characteristics through shared and non-shared generators.

Contribution

The paper presents a new GAN design that models multiple distributions with shared and non-shared generators to discover commonalities and particularities among them.

Findings

Effective modeling of multiple datasets like MNIST, CIFAR-10, CelebA.

Captures shared features across distributions.

Identifies unique aspects of each distribution.

Abstract

We propose a GAN design which models multiple distributions effectively and discovers their commonalities and particularities. Each data distribution is modeled with a mixture of $K$ generator distributions. As the generators are partially shared between the modeling of different true data distributions, shared ones captures the commonality of the distributions, while non-shared ones capture unique aspects of them. We show the effectiveness of our method on various datasets (MNIST, Fashion MNIST, CIFAR-10, Omniglot, CelebA) with compelling results.