Hierarchical Mixtures of Generators for Adversarial Learning

TL;DR

This paper introduces a hierarchical mixture of generators for GANs, enabling more diverse and high-quality sample generation by combining multiple local generators in a tree-structured hierarchy, trained via gradient descent.

Contribution

It proposes a novel hierarchical mixture model for GANs that improves sample diversity and quality, and allows for hierarchical knowledge extraction.

Findings

Achieved high-quality, diverse samples across five datasets.

The hierarchical structure facilitates interpretability and knowledge extraction.

Model trained effectively with gradient-based optimization.

Abstract

Generative adversarial networks (GANs) are deep neural networks that allow us to sample from an arbitrary probability distribution without explicitly estimating the distribution. There is a generator that takes a latent vector as input and transforms it into a valid sample from the distribution. There is also a discriminator that is trained to discriminate such fake samples from true samples of the distribution; at the same time, the generator is trained to generate fakes that the discriminator cannot tell apart from the true samples. Instead of learning a global generator, a recent approach involves training multiple generators each responsible from one part of the distribution. In this work, we review such approaches and propose the hierarchical mixture of generators, inspired from the hierarchical mixture of experts model, that learns a tree structure implementing a hierarchical clustering with soft splits in the decision nodes and local generators in the leaves. Since the generators are combined softly, the whole model is continuous and can be trained using gradient-based optimization, just like the original GAN model. Our experiments on five image data sets, namely, MNIST, FashionMNIST, UTZap50K, Oxford Flowers, and CelebA, show that our proposed model generates samples of high quality and diversity in terms of popular GAN evaluation metrics. The learned hierarchical structure also leads to knowledge extraction.