On GANs and GMMs

Abstract

A longstanding problem in machine learning is to find unsupervised methods that can learn the statistical structure of high dimensional signals. In recent years, GANs have gained much attention as a possible solution to the problem, and in particular have shown the ability to generate remarkably realistic high resolution sampled images. At the same time, many authors have pointed out that GANs may fail to model the full distribution ("mode collapse") and that using the learned models for anything other than generating samples may be very difficult. In this paper, we examine the utility of GANs in learning statistical models of images by comparing them to perhaps the simplest statistical model, the Gaussian Mixture Model. First, we present a simple method to evaluate generative models based on relative proportions of samples that fall into predetermined bins. Unlike previous automatic methods for evaluating models, our method does not rely on an additional neural network nor does it require approximating intractable computations. Second, we compare the performance of GANs to GMMs trained on the same datasets. While GMMs have previously been shown to be successful in modeling small patches of images, we show how to train them on full sized images despite the high dimensionality. Our results show that GMMs can generate realistic samples (although less sharp than those of GANs) but also capture the full distribution, which GANs fail to do. Furthermore, GMMs allow efficient inference and explicit representation of the underlying statistical structure. Finally, we discuss how GMMs can be used to generate sharp images.

arXiv commentAccepted to NIPS 2018

Cited by in corpus (20)

• DRIT++: Diverse Image-to-Image Translation via Disentangled Representations
• DiffWave: A Versatile Diffusion Model for Audio Synthesis
• Unconditional Audio Generation with Generative Adversarial Networks and Cycle Regularization
• A RAD approach to deep mixture models
• Top-k Training of GANs: Improving GAN Performance by Throwing Away Bad Samples
• Speaker Anonymization with Distribution-Preserving X-Vector Generation for the VoicePrivacy Challenge 2020
• Solving Linear Inverse Problems Using the Prior Implicit in a Denoiser
• GDPP: Learning Diverse Generations Using Determinantal Point Process
• MCL-GAN: Generative Adversarial Networks with Multiple Specialized Discriminators
• Mini-batch graphs for robust image classification
• A Bayes-Optimal View on Adversarial Examples
• Optimal Bayesian estimation of Gaussian mixtures with growing number of components
• Multimodal Image-to-Image Translation via Mutual Information Estimation and Maximization
• On Learning the Transformer Kernel
• A Benchmarking Initiative for Audio-Domain Music Generation Using the Freesound Loop Dataset
• IID-GAN: an IID Sampling Perspective for Regularizing Mode Collapse
• SetGAN: Improving the stability and diversity of generative models through a permutation invariant architecture
• Scenario generation for market risk models using generative neural networks
• Quantifying point cloud realism through adversarially learned latent representations
• Hierarchical Modes Exploring in Generative Adversarial Networks