Do GANs actually learn the distribution? An empirical study

Abstract

Do GANS (Generative Adversarial Nets) actually learn the target distribution? The foundational paper of (Goodfellow et al 2014) suggested they do, if they were given sufficiently large deep nets, sample size, and computation time. A recent theoretical analysis in Arora et al (to appear at ICML 2017) raised doubts whether the same holds when discriminator has finite size. It showed that the training objective can approach its optimum value even if the generated distribution has very low support ---in other words, the training objective is unable to prevent mode collapse. The current note reports experiments suggesting that such problems are not merely theoretical. It presents empirical evidence that well-known GANs approaches do learn distributions of fairly low support, and thus presumably are not learning the target distribution. The main technical contribution is a new proposed test, based upon the famous birthday paradox, for estimating the support size of the generated distribution.

References in corpus (1)

• Improved generator objectives for GANs

Cited by in corpus (17)

• Coulomb GANs: Provably Optimal Nash Equilibria via Potential Fields
• A black box for dark sector physics: Predicting dark matter annihilation feedback with conditional GANs
• A unified framework for 21cm tomography sample generation and parameter inference with Progressively Growing GANs
• GANGs: Generative Adversarial Network Games
• Theoretical limitations of Encoder-Decoder GAN architectures
• A New Semi-supervised Learning Benchmark for Classifying View and Diagnosing Aortic Stenosis from Echocardiograms
• Effect of The Latent Structure on Clustering with GANs
• Investigating Under and Overfitting in Wasserstein Generative Adversarial Networks
• Learning Shared Semantic Space with Correlation Alignment for Cross-modal Event Retrieval
• Orthogonal Wasserstein GANs
• Generative Models for Security: Attacks, Defenses, and Opportunities
• Graph-based Discriminators: Sample Complexity and Expressiveness
• Implicit Manifold Learning on Generative Adversarial Networks
• Detecting GAN generated errors
• To Regularize or Not To Regularize? The Bias Variance Trade-off in Regularized AEs
• Capturing the diversity of biological tuning curves using generative adversarial networks
• The Implicit Metropolis-Hastings Algorithm