Research map / Image generation / GANs
f-GAN: Training Generative Neural Samplers using Variational Divergence Minimization
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 |
|---|---|---|---|
| Use Pearson χ² rather than GAN's Jensen-Shannon-related objective (ablation) | f-GAN on MNIST · MNIST test | KDE mean log-likelihood (nats): 305 -> 429 | Helped |
| Use KL rather than GAN's Jensen-Shannon-related objective (ablation) | f-GAN on MNIST · MNIST test | KDE mean log-likelihood (nats): 305 -> 416 | Helped |
| Use reverse KL rather than GAN's Jensen-Shannon-related objective (ablation) | f-GAN on MNIST · MNIST test | KDE mean log-likelihood (nats): 305 -> 319 | Helped |
| Use squared Hellinger divergence rather than KL (ablation) | f-GAN on MNIST · MNIST test | KDE mean log-likelihood (nats): 416 -> -708 | Hurt |
| Use Jeffrey divergence rather than KL (ablation) | f-GAN on MNIST · MNIST test | KDE mean log-likelihood (nats): 416 -> -2101 | Hurt |
| Replace f-GAN with a variational autoencoder | MNIST generative modeling · MNIST test | KDE mean log-likelihood (nats): 416 -> 445 | Helped |
| Choose the divergence used for training (ablation) | Gaussian approximation of a mixture of Gaussians · Mixture of Gaussians | matching divergence objective: KL: 0.2808; KL-rev: 0.2414; JS: 0.1210; Jeffrey: 0.5236; Pearson: 0.648 | Helped |
| Use a different divergence to train a Gaussian when the model is misspecified (ablation) | f-GAN · Mixture of Gaussians | learned distribution fit: 0.2808; 0.2414; 0.1210; 0.5236; 0.648 | Mixed |
| Replace the original generator minimization term with maximizing generator output (ablation) | f-GAN · Training objective | stationary point preservation | No clear effect |
Closest papers by meaning
- Refining Deep Generative Models via Discriminator Gradient Flow (ICLR 2021 · similarity 0.89)
- Deep Generative Learning via Variational Gradient Flow (ICML 2019 · similarity 0.89)
- Adaptive Density Estimation for Generative Models (NeurIPS 2019 · similarity 0.89)
- Regularizing Generative Adversarial Networks Under Limited Data (CVPR 2021 · similarity 0.89)
- Precision-Recall Divergence Optimization for Generative Modeling with GANs and Normalizing Flows (NeurIPS 2023 · similarity 0.88)
- Composite Functional Gradient Learning of Generative Adversarial Models (ICML 2018 · similarity 0.88)
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.