Research map / Image generation / GANs
Fisher GAN
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 |
|---|---|---|---|
| Replace WGAN weight clipping with Fisher GAN (BN Uncond, 1000 labels) | Critic constraint · CIFAR-10 SSL 1000 labels | misclassification rate: 69.01 -> 36.42 | Helped |
| Use LN instead of BN in Fisher GAN critic (K+1 Uncond, 1000 labels) (ablation) | Fisher GAN K+1 · CIFAR-10 SSL 1000 labels | misclassification rate: 33.49 -> 19.74 | Helped |
| Add K+1 critic parametrization (LN, Uncond, 1000 labels) (ablation) | Fisher GAN LN · CIFAR-10 SSL 1000 labels | misclassification rate: 24.39 -> 19.74 | Helped |
| Add K+1 critic parametrization (LN, Cond, 1000 labels) (ablation) | Fisher GAN LN · CIFAR-10 SSL 1000 labels | misclassification rate: 26.78 -> 20.99 | Helped |
| Add K+1 critic parametrization (BN, Uncond, 1000 labels) (ablation) | Fisher GAN BN · CIFAR-10 SSL 1000 labels | misclassification rate: 36.42 -> 33.49 | Helped |
| Use conditional instead of unconditional generation (LN K+1, 1000 labels) (ablation) | Fisher GAN LN K+1 · CIFAR-10 SSL 1000 labels | misclassification rate: 19.74 -> 20.99 | Hurt |
| Use no normalization instead of LN (K+1 Uncond, 1000 labels) (ablation) | Fisher GAN K+1 · CIFAR-10 SSL 1000 labels | misclassification rate: 19.74 -> 21.15 | Hurt |
| Use no normalization instead of LN (K+1 Uncond, 8000 labels) (ablation) | Fisher GAN K+1 · CIFAR-10 SSL 8000 labels | misclassification rate: 14.81 -> 14.80 | No clear effect |
| Use Fisher GAN LN K+1 Uncond instead of Improved GAN FM | Semi-supervised GAN · CIFAR-10 SSL 8000 labels | misclassification rate: 17.72 -> 14.81 | Helped |
| Replace WGAN-GP with Fisher GAN ResNet (unsupervised) | ResNet GAN · CIFAR-10 unsupervised | inception score: 7.86 -> 7.90 | No clear effect |
| Replace WGAN-GP with Fisher GAN ResNet (supervised/labels) | ResNet GAN · CIFAR-10 supervised | inception score: 8.42 -> 8.16 | Hurt |
| Replace WGAN-GP with Fisher GAN in training speed/stability | ResNet GAN · CIFAR-10 inception vs wallclock | inception score | Helped |
Closest papers by meaning
- Adversarial Fisher Vectors for Unsupervised Representation Learning (NeurIPS 2019 · similarity 0.90)
- TRAINING GENERATIVE ADVERSARIAL NETWORKS VIA PRIMAL-DUAL SUBGRADIENT METHODS: A LAGRANGIAN PERSPECTIVE ON GAN (ICLR 2018 · similarity 0.87)
- On Stabilizing Generative Adversarial Training With Noise (CVPR 2019 · similarity 0.86)
- On-Line Adaptative Curriculum Learning for GANs (AAAI 2019 · similarity 0.86)
- MGAN: Training Generative Adversarial Nets with Multiple Generators (ICLR 2018 · similarity 0.86)
- MCGAN: Enhancing GAN Training with Regression-Based Generator Loss (AAAI 2025 · similarity 0.85)
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.