Research map / Optimization / Optimizers
Taming GANs with Lookahead-Minmax
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 Lookahead-minmax to AltGAN | AltGAN · CIFAR-10 | FID: 21.37 ± 1.60 -> 16.74 ± .46 | Helped |
| Add Lookahead-minmax to AltGAN | AltGAN · SVHN | FID: 7.84 ± 1.21 -> 3.87 ± .09 | Helped |
| Add Lookahead-minmax to ExtraGrad | ExtraGrad · CIFAR-10 | FID: 18.49 ± .99 -> 15.25 ± .30 | Helped |
| Add Lookahead-minmax to ExtraGrad | ExtraGrad · SVHN | FID: 4.08 ± .11 -> 3.20 ± .09 | Helped |
| Use nested rather than single Lookahead-minmax (ablation) | LA–AltGAN · ImageNet 32×32 | FID: 14.37 ± .20 -> 13.14 \pm .25 | Helped |
| Use nested rather than single Lookahead-minmax (ablation) | LA–ExtraGrad · ImageNet 32×32 | FID: 14.53 \pm .27 -> 15.05 \pm .96 | Mixed |
| Increase Lookahead k (ablation) | Lookahead-minmax GAN training · MNIST, CIFAR-10, SVHN, ImageNet | FID | Helped |
| Use EMA of weights (ablation) | LA–GAN · ImageNet 32×32 | FID: 14.37 \pm .20 -> 13.06\pm .20 | Helped |
| Average only the slow weights (ablation) | LA–AltGAN with EMA · ImageNet 32×32 | FID: 13.06\pm .20 -> 12.53\pm .06 | Helped |
| Use alpha=0.4 instead of alpha=0.5 (ablation) | LA–GDA on the full-batch bilinear game · Bilinear game | distance to optimum | Helped |
| Apply Lookahead-minmax rather than EMA-only averaging (ablation) | GAN training · CIFAR-10, SVHN, ImageNet | training stability | Helped |
Closest papers by meaning
- Lookahead Optimizer: k steps forward, 1 step back (NeurIPS 2019 · similarity 0.88)
- TRAINING GENERATIVE ADVERSARIAL NETWORKS VIA PRIMAL-DUAL SUBGRADIENT METHODS: A LAGRANGIAN PERSPECTIVE ON GAN (ICLR 2018 · similarity 0.88)
- Training Generative Adversarial Networks with Limited Data (NeurIPS 2020 · similarity 0.87)
- Hidden Convexity of Wasserstein GANs: Interpretable Generative Models with Closed-Form Solutions (ICLR 2022 · similarity 0.87)
- Improving Sharpness-Aware Minimization by Lookahead (ICML 2024 · similarity 0.87)
- SAN: Inducing Metrizability of GAN with Discriminative Normalized Linear Layer (ICLR 2024 · similarity 0.87)
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.