Research map / Optimization / Optimizers
The Numerics of GANs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add consensus regularization to both players | Simultaneous gradient ascent GAN training · Mixture of 8 Gaussians | Convergence | Helped |
| Add consensus regularization to both players | DC-GAN · CIFAR-10 | Inception score | Helped |
| Add consensus regularization to both players | DC-GAN without batch normalization · CIFAR-10 | Training stability | Helped |
| Add consensus regularization to both players | DC-GAN-like GAN · CelebA | Training stability | Helped |
| Add consensus regularization to both players | GAN with 4 convolutional layers · CIFAR-10 | Inception score | Helped |
| Add consensus regularization to both players | DC-GAN without batch normalization and with constant-width layers · GAN architectures tested in supplementary experiments | Training success | Helped |
| Use consensus regularization with different divergence objectives | DC-GAN without batch normalization and with constant-width layers · Standard GAN, Jensen-Shannon, and indicator divergences | Training success | Helped |
| Remove consensus regularization from the generator objective while retaining it for the discriminator (ablation) | Consensus optimization (smoothing optimizer) · DC-GAN architectures | Training performance | Hurt |
| Use discriminator-only consensus regularization (ablation) | Consensus optimization (smoothing optimizer) · DC-GAN with batch normalization | Training success | Mixed |
| Use discriminator-only consensus regularization (ablation) | Consensus optimization (smoothing optimizer) · DC-GAN with constant-width layers | Training success | Hurt |
| Add batch normalization to the standard DC-GAN architecture (ablation) | Consensus optimization · Standard DC-GAN | Training success | Hurt |
| Replace simultaneous gradient ascent with consensus optimization | GAN with additional RESNET layers · CIFAR-10 | Visual sample quality | Helped |
Closest papers by meaning
- An Online Learning Approach to Generative Adversarial Networks (ICLR 2018 · similarity 0.93)
- DO-GAN: A Double Oracle Framework for Generative Adversarial Networks (CVPR 2022 · similarity 0.92)
- Finding Mixed Nash Equilibria of Generative Adversarial Networks (ICML 2019 · similarity 0.91)
- Improving Generalization and Stability of Generative Adversarial Networks (ICLR 2019 · similarity 0.91)
- Are GANs Created Equal? A Large-Scale Study (NeurIPS 2018 · similarity 0.90)
- A Neural Tangent Kernel Perspective of GANs (ICML 2022 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.