mlsanityData to 7 Oct 2026

Do GANs actually learn the distribution? An empirical study

arXiv 1706.08224 ↗First posted 26 Jun 2017cs.LGcited by 23

Not on the map. Why?

Abstract (arXiv)

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.

Key facts. Cited by 23 papers in the map (data through 7 Oct 2026). First posted on arXiv 26 Jun 2017.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 6; 2022: 4; 2023: 4; 2024: 5; 2025: 3; 2026: 1.

23 papers.

Sort:

By area:

  1. 2023Physics of Language Models: Part 1, Learning Hierarchical Language Structurescited by 83
  2. 2021On Aliased Resizing and Surprising Subtleties in GAN Evaluationcited by 51
  3. 2022When do GANs replicate? On the choice of dataset sizecited by 34
  4. 2022Studying Bias in GANs through the Lens of Racecited by 31
  5. 2021Conditional Image Generation by Conditioning Variational Auto-Encoderscited by 25
  6. 2021A New Semi-supervised Learning Benchmark for Classifying View and Diagnosing Aortic Stenosis from Echocardiogramscited by 15
  7. 2023Synthetic Sample Selection for Generalized Zero-Shot Learningcited by 14
  8. 2023Ten Years of Generative Adversarial Nets (GANs): A survey of the state-of-the-artcited by 14
  9. 2021Learning stochastic object models from medical imaging measurements by use of advanced ambient generative adversarial networkscited by 12
  10. 2022RareGAN: Generating Samples for Rare Classescited by 9
  11. 2024An Inversion-based Measure of Memorization for Diffusion Modelscited by 6
  12. 2025Calibrating Generative Models to Distributional Constraintscited by 5
11 more
  1. 2021Black-Box Diagnosis and Calibration on GAN Intra-Mode Collapse: A Pilot Studycited by 3
  2. 2021Probabilistic Autoencoder using Fisher Informationcited by 3
  3. 2023On quantifying and improving realism of images generated with diffusioncited by 3
  4. 2024Training Implicit Generative Models via an Invariant Statistical Losscited by 3
  5. 2024Ai-Sampler: Adversarial Learning of Markov kernels with involutive mapscited by 1
  6. 2022Time Efficient Training of Progressive Generative Adversarial Network using Depthwise Separable Convolution and Super Resolution Generative Adversarial Networkcited by 0
  7. 2024DE-CGAN: Boosting rTMS Treatment Prediction with Diversity Enhancing Conditional Generative Adversarial Networkscited by 0
  8. 2024Enhancing GANs with Contrastive Learning-Based Multistage Progressive Finetuning SNN and RL-Based External Optimizationcited by 0
  9. 2025Bringing Balance to Hand Shape Classification: Mitigating Data Imbalance Through Generative Modelscited by 0
  10. 2025Learning Latent Representations for Image Translation using Frequency Distributed CycleGANcited by 0
  11. 2026Fine-Tuning Fixes Mode Collapse and Over-Dispersion in LLMscited by 0